1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_capsicum.h" 43 #include "opt_ddb.h" 44 #include "opt_ktrace.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 49 #include <sys/capsicum.h> 50 #include <sys/conf.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/filio.h> 55 #include <sys/jail.h> 56 #include <sys/kernel.h> 57 #include <sys/limits.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mount.h> 61 #include <sys/mutex.h> 62 #include <sys/namei.h> 63 #include <sys/selinfo.h> 64 #include <sys/poll.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/protosw.h> 68 #include <sys/racct.h> 69 #include <sys/resourcevar.h> 70 #include <sys/sbuf.h> 71 #include <sys/signalvar.h> 72 #include <sys/kdb.h> 73 #include <sys/smr.h> 74 #include <sys/stat.h> 75 #include <sys/sx.h> 76 #include <sys/syscallsubr.h> 77 #include <sys/sysctl.h> 78 #include <sys/sysproto.h> 79 #include <sys/unistd.h> 80 #include <sys/user.h> 81 #include <sys/vnode.h> 82 #include <sys/ktrace.h> 83 84 #include <net/vnet.h> 85 86 #include <security/audit/audit.h> 87 88 #include <vm/uma.h> 89 #include <vm/vm.h> 90 91 #include <ddb/ddb.h> 92 93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes"); 95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors"); 96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 97 "file desc to leader structures"); 98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 100 101 MALLOC_DECLARE(M_FADVISE); 102 103 static __read_mostly uma_zone_t file_zone; 104 static __read_mostly uma_zone_t filedesc0_zone; 105 __read_mostly uma_zone_t pwd_zone; 106 VFS_SMR_DECLARE; 107 108 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 109 struct thread *td, bool holdleaders, bool audit); 110 static void export_file_to_kinfo(struct file *fp, int fd, 111 cap_rights_t *rightsp, struct kinfo_file *kif, 112 struct filedesc *fdp, int flags); 113 static int fd_first_free(struct filedesc *fdp, int low, int size); 114 static void fdgrowtable(struct filedesc *fdp, int nfd); 115 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 116 static void fdunused(struct filedesc *fdp, int fd); 117 static void fdused(struct filedesc *fdp, int fd); 118 static int fget_unlocked_seq(struct thread *td, int fd, 119 cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp); 120 static int getmaxfd(struct thread *td); 121 static u_long *filecaps_copy_prep(const struct filecaps *src); 122 static void filecaps_copy_finish(const struct filecaps *src, 123 struct filecaps *dst, u_long *ioctls); 124 static u_long *filecaps_free_prep(struct filecaps *fcaps); 125 static void filecaps_free_finish(u_long *ioctls); 126 127 static struct pwd *pwd_alloc(void); 128 129 /* 130 * Each process has: 131 * 132 * - An array of open file descriptors (fd_ofiles) 133 * - An array of file flags (fd_ofileflags) 134 * - A bitmap recording which descriptors are in use (fd_map) 135 * 136 * A process starts out with NDFILE descriptors. The value of NDFILE has 137 * been selected based the historical limit of 20 open files, and an 138 * assumption that the majority of processes, especially short-lived 139 * processes like shells, will never need more. 140 * 141 * If this initial allocation is exhausted, a larger descriptor table and 142 * map are allocated dynamically, and the pointers in the process's struct 143 * filedesc are updated to point to those. This is repeated every time 144 * the process runs out of file descriptors (provided it hasn't hit its 145 * resource limit). 146 * 147 * Since threads may hold references to individual descriptor table 148 * entries, the tables are never freed. Instead, they are placed on a 149 * linked list and freed only when the struct filedesc is released. 150 */ 151 #define NDFILE 20 152 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 153 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 154 #define NDSLOT(x) ((x) / NDENTRIES) 155 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 156 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 157 158 #define FILEDESC_FOREACH_FDE(fdp, _iterator, _fde) \ 159 struct filedesc *_fdp = (fdp); \ 160 int _lastfile = fdlastfile_single(_fdp); \ 161 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 162 if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL) 163 164 #define FILEDESC_FOREACH_FP(fdp, _iterator, _fp) \ 165 struct filedesc *_fdp = (fdp); \ 166 int _lastfile = fdlastfile_single(_fdp); \ 167 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \ 168 if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL) 169 170 /* 171 * SLIST entry used to keep track of ofiles which must be reclaimed when 172 * the process exits. 173 */ 174 struct freetable { 175 struct fdescenttbl *ft_table; 176 SLIST_ENTRY(freetable) ft_next; 177 }; 178 179 /* 180 * Initial allocation: a filedesc structure + the head of SLIST used to 181 * keep track of old ofiles + enough space for NDFILE descriptors. 182 */ 183 184 struct fdescenttbl0 { 185 int fdt_nfiles; 186 struct filedescent fdt_ofiles[NDFILE]; 187 }; 188 189 struct filedesc0 { 190 struct filedesc fd_fd; 191 SLIST_HEAD(, freetable) fd_free; 192 struct fdescenttbl0 fd_dfiles; 193 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 194 }; 195 196 /* 197 * Descriptor management. 198 */ 199 static int __exclusive_cache_line openfiles; /* actual number of open files */ 200 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 201 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 202 203 /* 204 * If low >= size, just return low. Otherwise find the first zero bit in the 205 * given bitmap, starting at low and not exceeding size - 1. Return size if 206 * not found. 207 */ 208 static int 209 fd_first_free(struct filedesc *fdp, int low, int size) 210 { 211 NDSLOTTYPE *map = fdp->fd_map; 212 NDSLOTTYPE mask; 213 int off, maxoff; 214 215 if (low >= size) 216 return (low); 217 218 off = NDSLOT(low); 219 if (low % NDENTRIES) { 220 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 221 if ((mask &= ~map[off]) != 0UL) 222 return (off * NDENTRIES + ffsl(mask) - 1); 223 ++off; 224 } 225 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 226 if (map[off] != ~0UL) 227 return (off * NDENTRIES + ffsl(~map[off]) - 1); 228 return (size); 229 } 230 231 /* 232 * Find the last used fd. 233 * 234 * Call this variant if fdp can't be modified by anyone else (e.g, during exec). 235 * Otherwise use fdlastfile. 236 */ 237 int 238 fdlastfile_single(struct filedesc *fdp) 239 { 240 NDSLOTTYPE *map = fdp->fd_map; 241 int off, minoff; 242 243 off = NDSLOT(fdp->fd_nfiles - 1); 244 for (minoff = NDSLOT(0); off >= minoff; --off) 245 if (map[off] != 0) 246 return (off * NDENTRIES + flsl(map[off]) - 1); 247 return (-1); 248 } 249 250 int 251 fdlastfile(struct filedesc *fdp) 252 { 253 254 FILEDESC_LOCK_ASSERT(fdp); 255 return (fdlastfile_single(fdp)); 256 } 257 258 static int 259 fdisused(struct filedesc *fdp, int fd) 260 { 261 262 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 263 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 264 265 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 266 } 267 268 /* 269 * Mark a file descriptor as used. 270 */ 271 static void 272 fdused_init(struct filedesc *fdp, int fd) 273 { 274 275 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 276 277 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 278 } 279 280 static void 281 fdused(struct filedesc *fdp, int fd) 282 { 283 284 FILEDESC_XLOCK_ASSERT(fdp); 285 286 fdused_init(fdp, fd); 287 if (fd == fdp->fd_freefile) 288 fdp->fd_freefile++; 289 } 290 291 /* 292 * Mark a file descriptor as unused. 293 */ 294 static void 295 fdunused(struct filedesc *fdp, int fd) 296 { 297 298 FILEDESC_XLOCK_ASSERT(fdp); 299 300 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 301 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 302 ("fd=%d is still in use", fd)); 303 304 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 305 if (fd < fdp->fd_freefile) 306 fdp->fd_freefile = fd; 307 } 308 309 /* 310 * Free a file descriptor. 311 * 312 * Avoid some work if fdp is about to be destroyed. 313 */ 314 static inline void 315 fdefree_last(struct filedescent *fde) 316 { 317 318 filecaps_free(&fde->fde_caps); 319 } 320 321 static inline void 322 fdfree(struct filedesc *fdp, int fd) 323 { 324 struct filedescent *fde; 325 326 FILEDESC_XLOCK_ASSERT(fdp); 327 fde = &fdp->fd_ofiles[fd]; 328 #ifdef CAPABILITIES 329 seqc_write_begin(&fde->fde_seqc); 330 #endif 331 fde->fde_file = NULL; 332 #ifdef CAPABILITIES 333 seqc_write_end(&fde->fde_seqc); 334 #endif 335 fdefree_last(fde); 336 fdunused(fdp, fd); 337 } 338 339 /* 340 * System calls on descriptors. 341 */ 342 #ifndef _SYS_SYSPROTO_H_ 343 struct getdtablesize_args { 344 int dummy; 345 }; 346 #endif 347 /* ARGSUSED */ 348 int 349 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 350 { 351 #ifdef RACCT 352 uint64_t lim; 353 #endif 354 355 td->td_retval[0] = getmaxfd(td); 356 #ifdef RACCT 357 PROC_LOCK(td->td_proc); 358 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 359 PROC_UNLOCK(td->td_proc); 360 if (lim < td->td_retval[0]) 361 td->td_retval[0] = lim; 362 #endif 363 return (0); 364 } 365 366 /* 367 * Duplicate a file descriptor to a particular value. 368 * 369 * Note: keep in mind that a potential race condition exists when closing 370 * descriptors from a shared descriptor table (via rfork). 371 */ 372 #ifndef _SYS_SYSPROTO_H_ 373 struct dup2_args { 374 u_int from; 375 u_int to; 376 }; 377 #endif 378 /* ARGSUSED */ 379 int 380 sys_dup2(struct thread *td, struct dup2_args *uap) 381 { 382 383 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); 384 } 385 386 /* 387 * Duplicate a file descriptor. 388 */ 389 #ifndef _SYS_SYSPROTO_H_ 390 struct dup_args { 391 u_int fd; 392 }; 393 #endif 394 /* ARGSUSED */ 395 int 396 sys_dup(struct thread *td, struct dup_args *uap) 397 { 398 399 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); 400 } 401 402 /* 403 * The file control system call. 404 */ 405 #ifndef _SYS_SYSPROTO_H_ 406 struct fcntl_args { 407 int fd; 408 int cmd; 409 long arg; 410 }; 411 #endif 412 /* ARGSUSED */ 413 int 414 sys_fcntl(struct thread *td, struct fcntl_args *uap) 415 { 416 417 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); 418 } 419 420 int 421 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) 422 { 423 struct flock fl; 424 struct __oflock ofl; 425 intptr_t arg1; 426 int error, newcmd; 427 428 error = 0; 429 newcmd = cmd; 430 switch (cmd) { 431 case F_OGETLK: 432 case F_OSETLK: 433 case F_OSETLKW: 434 /* 435 * Convert old flock structure to new. 436 */ 437 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); 438 fl.l_start = ofl.l_start; 439 fl.l_len = ofl.l_len; 440 fl.l_pid = ofl.l_pid; 441 fl.l_type = ofl.l_type; 442 fl.l_whence = ofl.l_whence; 443 fl.l_sysid = 0; 444 445 switch (cmd) { 446 case F_OGETLK: 447 newcmd = F_GETLK; 448 break; 449 case F_OSETLK: 450 newcmd = F_SETLK; 451 break; 452 case F_OSETLKW: 453 newcmd = F_SETLKW; 454 break; 455 } 456 arg1 = (intptr_t)&fl; 457 break; 458 case F_GETLK: 459 case F_SETLK: 460 case F_SETLKW: 461 case F_SETLK_REMOTE: 462 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); 463 arg1 = (intptr_t)&fl; 464 break; 465 default: 466 arg1 = arg; 467 break; 468 } 469 if (error) 470 return (error); 471 error = kern_fcntl(td, fd, newcmd, arg1); 472 if (error) 473 return (error); 474 if (cmd == F_OGETLK) { 475 ofl.l_start = fl.l_start; 476 ofl.l_len = fl.l_len; 477 ofl.l_pid = fl.l_pid; 478 ofl.l_type = fl.l_type; 479 ofl.l_whence = fl.l_whence; 480 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); 481 } else if (cmd == F_GETLK) { 482 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); 483 } 484 return (error); 485 } 486 487 int 488 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 489 { 490 struct filedesc *fdp; 491 struct flock *flp; 492 struct file *fp, *fp2; 493 struct filedescent *fde; 494 struct proc *p; 495 struct vnode *vp; 496 struct mount *mp; 497 struct kinfo_file *kif; 498 int error, flg, kif_sz, seals, tmp; 499 uint64_t bsize; 500 off_t foffset; 501 502 error = 0; 503 flg = F_POSIX; 504 p = td->td_proc; 505 fdp = p->p_fd; 506 507 AUDIT_ARG_FD(cmd); 508 AUDIT_ARG_CMD(cmd); 509 switch (cmd) { 510 case F_DUPFD: 511 tmp = arg; 512 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); 513 break; 514 515 case F_DUPFD_CLOEXEC: 516 tmp = arg; 517 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); 518 break; 519 520 case F_DUP2FD: 521 tmp = arg; 522 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); 523 break; 524 525 case F_DUP2FD_CLOEXEC: 526 tmp = arg; 527 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); 528 break; 529 530 case F_GETFD: 531 error = EBADF; 532 FILEDESC_SLOCK(fdp); 533 fde = fdeget_noref(fdp, fd); 534 if (fde != NULL) { 535 td->td_retval[0] = 536 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 537 error = 0; 538 } 539 FILEDESC_SUNLOCK(fdp); 540 break; 541 542 case F_SETFD: 543 error = EBADF; 544 FILEDESC_XLOCK(fdp); 545 fde = fdeget_noref(fdp, fd); 546 if (fde != NULL) { 547 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 548 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 549 error = 0; 550 } 551 FILEDESC_XUNLOCK(fdp); 552 break; 553 554 case F_GETFL: 555 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); 556 if (error != 0) 557 break; 558 td->td_retval[0] = OFLAGS(fp->f_flag); 559 fdrop(fp, td); 560 break; 561 562 case F_SETFL: 563 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); 564 if (error != 0) 565 break; 566 if (fp->f_ops == &path_fileops) { 567 fdrop(fp, td); 568 error = EBADF; 569 break; 570 } 571 do { 572 tmp = flg = fp->f_flag; 573 tmp &= ~FCNTLFLAGS; 574 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 575 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 576 tmp = fp->f_flag & FNONBLOCK; 577 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 578 if (error != 0) { 579 fdrop(fp, td); 580 break; 581 } 582 tmp = fp->f_flag & FASYNC; 583 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 584 if (error == 0) { 585 fdrop(fp, td); 586 break; 587 } 588 atomic_clear_int(&fp->f_flag, FNONBLOCK); 589 tmp = 0; 590 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 591 fdrop(fp, td); 592 break; 593 594 case F_GETOWN: 595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); 596 if (error != 0) 597 break; 598 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 599 if (error == 0) 600 td->td_retval[0] = tmp; 601 fdrop(fp, td); 602 break; 603 604 case F_SETOWN: 605 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); 606 if (error != 0) 607 break; 608 tmp = arg; 609 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 610 fdrop(fp, td); 611 break; 612 613 case F_SETLK_REMOTE: 614 error = priv_check(td, PRIV_NFS_LOCKD); 615 if (error != 0) 616 return (error); 617 flg = F_REMOTE; 618 goto do_setlk; 619 620 case F_SETLKW: 621 flg |= F_WAIT; 622 /* FALLTHROUGH F_SETLK */ 623 624 case F_SETLK: 625 do_setlk: 626 flp = (struct flock *)arg; 627 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { 628 error = EINVAL; 629 break; 630 } 631 632 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 633 if (error != 0) 634 break; 635 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 636 error = EBADF; 637 fdrop(fp, td); 638 break; 639 } 640 641 if (flp->l_whence == SEEK_CUR) { 642 foffset = foffset_get(fp); 643 if (foffset < 0 || 644 (flp->l_start > 0 && 645 foffset > OFF_MAX - flp->l_start)) { 646 error = EOVERFLOW; 647 fdrop(fp, td); 648 break; 649 } 650 flp->l_start += foffset; 651 } 652 653 vp = fp->f_vnode; 654 switch (flp->l_type) { 655 case F_RDLCK: 656 if ((fp->f_flag & FREAD) == 0) { 657 error = EBADF; 658 break; 659 } 660 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 661 PROC_LOCK(p->p_leader); 662 p->p_leader->p_flag |= P_ADVLOCK; 663 PROC_UNLOCK(p->p_leader); 664 } 665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 666 flp, flg); 667 break; 668 case F_WRLCK: 669 if ((fp->f_flag & FWRITE) == 0) { 670 error = EBADF; 671 break; 672 } 673 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { 674 PROC_LOCK(p->p_leader); 675 p->p_leader->p_flag |= P_ADVLOCK; 676 PROC_UNLOCK(p->p_leader); 677 } 678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 679 flp, flg); 680 break; 681 case F_UNLCK: 682 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 683 flp, flg); 684 break; 685 case F_UNLCKSYS: 686 if (flg != F_REMOTE) { 687 error = EINVAL; 688 break; 689 } 690 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 691 F_UNLCKSYS, flp, flg); 692 break; 693 default: 694 error = EINVAL; 695 break; 696 } 697 if (error != 0 || flp->l_type == F_UNLCK || 698 flp->l_type == F_UNLCKSYS) { 699 fdrop(fp, td); 700 break; 701 } 702 703 /* 704 * Check for a race with close. 705 * 706 * The vnode is now advisory locked (or unlocked, but this case 707 * is not really important) as the caller requested. 708 * We had to drop the filedesc lock, so we need to recheck if 709 * the descriptor is still valid, because if it was closed 710 * in the meantime we need to remove advisory lock from the 711 * vnode - close on any descriptor leading to an advisory 712 * locked vnode, removes that lock. 713 * We will return 0 on purpose in that case, as the result of 714 * successful advisory lock might have been externally visible 715 * already. This is fine - effectively we pretend to the caller 716 * that the closing thread was a bit slower and that the 717 * advisory lock succeeded before the close. 718 */ 719 error = fget_unlocked(td, fd, &cap_no_rights, &fp2); 720 if (error != 0) { 721 fdrop(fp, td); 722 break; 723 } 724 if (fp != fp2) { 725 flp->l_whence = SEEK_SET; 726 flp->l_start = 0; 727 flp->l_len = 0; 728 flp->l_type = F_UNLCK; 729 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 730 F_UNLCK, flp, F_POSIX); 731 } 732 fdrop(fp, td); 733 fdrop(fp2, td); 734 break; 735 736 case F_GETLK: 737 error = fget_unlocked(td, fd, &cap_flock_rights, &fp); 738 if (error != 0) 739 break; 740 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 741 error = EBADF; 742 fdrop(fp, td); 743 break; 744 } 745 flp = (struct flock *)arg; 746 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 747 flp->l_type != F_UNLCK) { 748 error = EINVAL; 749 fdrop(fp, td); 750 break; 751 } 752 if (flp->l_whence == SEEK_CUR) { 753 foffset = foffset_get(fp); 754 if ((flp->l_start > 0 && 755 foffset > OFF_MAX - flp->l_start) || 756 (flp->l_start < 0 && 757 foffset < OFF_MIN - flp->l_start)) { 758 error = EOVERFLOW; 759 fdrop(fp, td); 760 break; 761 } 762 flp->l_start += foffset; 763 } 764 vp = fp->f_vnode; 765 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 766 F_POSIX); 767 fdrop(fp, td); 768 break; 769 770 case F_ADD_SEALS: 771 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 772 if (error != 0) 773 break; 774 error = fo_add_seals(fp, arg); 775 fdrop(fp, td); 776 break; 777 778 case F_GET_SEALS: 779 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 780 if (error != 0) 781 break; 782 if (fo_get_seals(fp, &seals) == 0) 783 td->td_retval[0] = seals; 784 else 785 error = EINVAL; 786 fdrop(fp, td); 787 break; 788 789 case F_RDAHEAD: 790 arg = arg ? 128 * 1024: 0; 791 /* FALLTHROUGH */ 792 case F_READAHEAD: 793 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 794 if (error != 0) 795 break; 796 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) { 797 fdrop(fp, td); 798 error = EBADF; 799 break; 800 } 801 vp = fp->f_vnode; 802 if (vp->v_type != VREG) { 803 fdrop(fp, td); 804 error = ENOTTY; 805 break; 806 } 807 808 /* 809 * Exclusive lock synchronizes against f_seqcount reads and 810 * writes in sequential_heuristic(). 811 */ 812 error = vn_lock(vp, LK_EXCLUSIVE); 813 if (error != 0) { 814 fdrop(fp, td); 815 break; 816 } 817 if (arg >= 0) { 818 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 819 arg = MIN(arg, INT_MAX - bsize + 1); 820 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, 821 (arg + bsize - 1) / bsize); 822 atomic_set_int(&fp->f_flag, FRDAHEAD); 823 } else { 824 atomic_clear_int(&fp->f_flag, FRDAHEAD); 825 } 826 VOP_UNLOCK(vp); 827 fdrop(fp, td); 828 break; 829 830 case F_ISUNIONSTACK: 831 /* 832 * Check if the vnode is part of a union stack (either the 833 * "union" flag from mount(2) or unionfs). 834 * 835 * Prior to introduction of this op libc's readdir would call 836 * fstatfs(2), in effect unnecessarily copying kilobytes of 837 * data just to check fs name and a mount flag. 838 * 839 * Fixing the code to handle everything in the kernel instead 840 * is a non-trivial endeavor and has low priority, thus this 841 * horrible kludge facilitates the current behavior in a much 842 * cheaper manner until someone(tm) sorts this out. 843 */ 844 error = fget_unlocked(td, fd, &cap_no_rights, &fp); 845 if (error != 0) 846 break; 847 if (fp->f_type != DTYPE_VNODE) { 848 fdrop(fp, td); 849 error = EBADF; 850 break; 851 } 852 vp = fp->f_vnode; 853 /* 854 * Since we don't prevent dooming the vnode even non-null mp 855 * found can become immediately stale. This is tolerable since 856 * mount points are type-stable (providing safe memory access) 857 * and any vfs op on this vnode going forward will return an 858 * error (meaning return value in this case is meaningless). 859 */ 860 mp = atomic_load_ptr(&vp->v_mount); 861 if (__predict_false(mp == NULL)) { 862 fdrop(fp, td); 863 error = EBADF; 864 break; 865 } 866 td->td_retval[0] = 0; 867 if (mp->mnt_kern_flag & MNTK_UNIONFS || 868 mp->mnt_flag & MNT_UNION) 869 td->td_retval[0] = 1; 870 fdrop(fp, td); 871 break; 872 873 case F_KINFO: 874 #ifdef CAPABILITY_MODE 875 if (IN_CAPABILITY_MODE(td)) { 876 error = ECAPMODE; 877 break; 878 } 879 #endif 880 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz)); 881 if (error != 0) 882 break; 883 if (kif_sz != sizeof(*kif)) { 884 error = EINVAL; 885 break; 886 } 887 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO); 888 FILEDESC_SLOCK(fdp); 889 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL); 890 if (error == 0 && fhold(fp)) { 891 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0); 892 FILEDESC_SUNLOCK(fdp); 893 fdrop(fp, td); 894 if ((kif->kf_status & KF_ATTR_VALID) != 0) { 895 kif->kf_structsize = sizeof(*kif); 896 error = copyout(kif, (void *)arg, sizeof(*kif)); 897 } else { 898 error = EBADF; 899 } 900 } else { 901 FILEDESC_SUNLOCK(fdp); 902 if (error == 0) 903 error = EBADF; 904 } 905 free(kif, M_TEMP); 906 break; 907 908 default: 909 error = EINVAL; 910 break; 911 } 912 return (error); 913 } 914 915 static int 916 getmaxfd(struct thread *td) 917 { 918 919 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); 920 } 921 922 /* 923 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 924 */ 925 int 926 kern_dup(struct thread *td, u_int mode, int flags, int old, int new) 927 { 928 struct filedesc *fdp; 929 struct filedescent *oldfde, *newfde; 930 struct proc *p; 931 struct file *delfp, *oldfp; 932 u_long *oioctls, *nioctls; 933 int error, maxfd; 934 935 p = td->td_proc; 936 fdp = p->p_fd; 937 oioctls = NULL; 938 939 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); 940 MPASS(mode < FDDUP_LASTMODE); 941 942 AUDIT_ARG_FD(old); 943 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ 944 945 /* 946 * Verify we have a valid descriptor to dup from and possibly to 947 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 948 * return EINVAL when the new descriptor is out of bounds. 949 */ 950 if (old < 0) 951 return (EBADF); 952 if (new < 0) 953 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 954 maxfd = getmaxfd(td); 955 if (new >= maxfd) 956 return (mode == FDDUP_FCNTL ? EINVAL : EBADF); 957 958 error = EBADF; 959 FILEDESC_XLOCK(fdp); 960 if (fget_noref(fdp, old) == NULL) 961 goto unlock; 962 if (mode == FDDUP_FIXED && old == new) { 963 td->td_retval[0] = new; 964 if (flags & FDDUP_FLAG_CLOEXEC) 965 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 966 error = 0; 967 goto unlock; 968 } 969 970 oldfde = &fdp->fd_ofiles[old]; 971 oldfp = oldfde->fde_file; 972 if (!fhold(oldfp)) 973 goto unlock; 974 975 /* 976 * If the caller specified a file descriptor, make sure the file 977 * table is large enough to hold it, and grab it. Otherwise, just 978 * allocate a new descriptor the usual way. 979 */ 980 switch (mode) { 981 case FDDUP_NORMAL: 982 case FDDUP_FCNTL: 983 if ((error = fdalloc(td, new, &new)) != 0) { 984 fdrop(oldfp, td); 985 goto unlock; 986 } 987 break; 988 case FDDUP_FIXED: 989 if (new >= fdp->fd_nfiles) { 990 /* 991 * The resource limits are here instead of e.g. 992 * fdalloc(), because the file descriptor table may be 993 * shared between processes, so we can't really use 994 * racct_add()/racct_sub(). Instead of counting the 995 * number of actually allocated descriptors, just put 996 * the limit on the size of the file descriptor table. 997 */ 998 #ifdef RACCT 999 if (RACCT_ENABLED()) { 1000 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); 1001 if (error != 0) { 1002 error = EMFILE; 1003 fdrop(oldfp, td); 1004 goto unlock; 1005 } 1006 } 1007 #endif 1008 fdgrowtable_exp(fdp, new + 1); 1009 } 1010 if (!fdisused(fdp, new)) 1011 fdused(fdp, new); 1012 break; 1013 default: 1014 KASSERT(0, ("%s unsupported mode %d", __func__, mode)); 1015 } 1016 1017 KASSERT(old != new, ("new fd is same as old")); 1018 1019 /* Refetch oldfde because the table may have grown and old one freed. */ 1020 oldfde = &fdp->fd_ofiles[old]; 1021 KASSERT(oldfp == oldfde->fde_file, 1022 ("fdt_ofiles shift from growth observed at fd %d", 1023 old)); 1024 1025 newfde = &fdp->fd_ofiles[new]; 1026 delfp = newfde->fde_file; 1027 1028 nioctls = filecaps_copy_prep(&oldfde->fde_caps); 1029 1030 /* 1031 * Duplicate the source descriptor. 1032 */ 1033 #ifdef CAPABILITIES 1034 seqc_write_begin(&newfde->fde_seqc); 1035 #endif 1036 oioctls = filecaps_free_prep(&newfde->fde_caps); 1037 fde_copy(oldfde, newfde); 1038 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 1039 nioctls); 1040 if ((flags & FDDUP_FLAG_CLOEXEC) != 0) 1041 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 1042 else 1043 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 1044 #ifdef CAPABILITIES 1045 seqc_write_end(&newfde->fde_seqc); 1046 #endif 1047 td->td_retval[0] = new; 1048 1049 error = 0; 1050 1051 if (delfp != NULL) { 1052 (void) closefp(fdp, new, delfp, td, true, false); 1053 FILEDESC_UNLOCK_ASSERT(fdp); 1054 } else { 1055 unlock: 1056 FILEDESC_XUNLOCK(fdp); 1057 } 1058 1059 filecaps_free_finish(oioctls); 1060 return (error); 1061 } 1062 1063 static void 1064 sigiofree(struct sigio *sigio) 1065 { 1066 crfree(sigio->sio_ucred); 1067 free(sigio, M_SIGIO); 1068 } 1069 1070 static struct sigio * 1071 funsetown_locked(struct sigio *sigio) 1072 { 1073 struct proc *p; 1074 struct pgrp *pg; 1075 1076 SIGIO_ASSERT_LOCKED(); 1077 1078 if (sigio == NULL) 1079 return (NULL); 1080 *sigio->sio_myref = NULL; 1081 if (sigio->sio_pgid < 0) { 1082 pg = sigio->sio_pgrp; 1083 PGRP_LOCK(pg); 1084 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); 1085 PGRP_UNLOCK(pg); 1086 } else { 1087 p = sigio->sio_proc; 1088 PROC_LOCK(p); 1089 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); 1090 PROC_UNLOCK(p); 1091 } 1092 return (sigio); 1093 } 1094 1095 /* 1096 * If sigio is on the list associated with a process or process group, 1097 * disable signalling from the device, remove sigio from the list and 1098 * free sigio. 1099 */ 1100 void 1101 funsetown(struct sigio **sigiop) 1102 { 1103 struct sigio *sigio; 1104 1105 /* Racy check, consumers must provide synchronization. */ 1106 if (*sigiop == NULL) 1107 return; 1108 1109 SIGIO_LOCK(); 1110 sigio = funsetown_locked(*sigiop); 1111 SIGIO_UNLOCK(); 1112 if (sigio != NULL) 1113 sigiofree(sigio); 1114 } 1115 1116 /* 1117 * Free a list of sigio structures. The caller must ensure that new sigio 1118 * structures cannot be added after this point. For process groups this is 1119 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves 1120 * as an interlock. 1121 */ 1122 void 1123 funsetownlst(struct sigiolst *sigiolst) 1124 { 1125 struct proc *p; 1126 struct pgrp *pg; 1127 struct sigio *sigio, *tmp; 1128 1129 /* Racy check. */ 1130 sigio = SLIST_FIRST(sigiolst); 1131 if (sigio == NULL) 1132 return; 1133 1134 p = NULL; 1135 pg = NULL; 1136 1137 SIGIO_LOCK(); 1138 sigio = SLIST_FIRST(sigiolst); 1139 if (sigio == NULL) { 1140 SIGIO_UNLOCK(); 1141 return; 1142 } 1143 1144 /* 1145 * Every entry of the list should belong to a single proc or pgrp. 1146 */ 1147 if (sigio->sio_pgid < 0) { 1148 pg = sigio->sio_pgrp; 1149 sx_assert(&proctree_lock, SX_XLOCKED); 1150 PGRP_LOCK(pg); 1151 } else /* if (sigio->sio_pgid > 0) */ { 1152 p = sigio->sio_proc; 1153 PROC_LOCK(p); 1154 KASSERT((p->p_flag & P_WEXIT) != 0, 1155 ("%s: process %p is not exiting", __func__, p)); 1156 } 1157 1158 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) { 1159 *sigio->sio_myref = NULL; 1160 if (pg != NULL) { 1161 KASSERT(sigio->sio_pgid < 0, 1162 ("Proc sigio in pgrp sigio list")); 1163 KASSERT(sigio->sio_pgrp == pg, 1164 ("Bogus pgrp in sigio list")); 1165 } else /* if (p != NULL) */ { 1166 KASSERT(sigio->sio_pgid > 0, 1167 ("Pgrp sigio in proc sigio list")); 1168 KASSERT(sigio->sio_proc == p, 1169 ("Bogus proc in sigio list")); 1170 } 1171 } 1172 1173 if (pg != NULL) 1174 PGRP_UNLOCK(pg); 1175 else 1176 PROC_UNLOCK(p); 1177 SIGIO_UNLOCK(); 1178 1179 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp) 1180 sigiofree(sigio); 1181 } 1182 1183 /* 1184 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1185 * 1186 * After permission checking, add a sigio structure to the sigio list for 1187 * the process or process group. 1188 */ 1189 int 1190 fsetown(pid_t pgid, struct sigio **sigiop) 1191 { 1192 struct proc *proc; 1193 struct pgrp *pgrp; 1194 struct sigio *osigio, *sigio; 1195 int ret; 1196 1197 if (pgid == 0) { 1198 funsetown(sigiop); 1199 return (0); 1200 } 1201 1202 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1203 sigio->sio_pgid = pgid; 1204 sigio->sio_ucred = crhold(curthread->td_ucred); 1205 sigio->sio_myref = sigiop; 1206 1207 ret = 0; 1208 if (pgid > 0) { 1209 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc); 1210 SIGIO_LOCK(); 1211 osigio = funsetown_locked(*sigiop); 1212 if (ret == 0) { 1213 PROC_LOCK(proc); 1214 _PRELE(proc); 1215 if ((proc->p_flag & P_WEXIT) != 0) { 1216 ret = ESRCH; 1217 } else if (proc->p_session != 1218 curthread->td_proc->p_session) { 1219 /* 1220 * Policy - Don't allow a process to FSETOWN a 1221 * process in another session. 1222 * 1223 * Remove this test to allow maximum flexibility 1224 * or restrict FSETOWN to the current process or 1225 * process group for maximum safety. 1226 */ 1227 ret = EPERM; 1228 } else { 1229 sigio->sio_proc = proc; 1230 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, 1231 sio_pgsigio); 1232 } 1233 PROC_UNLOCK(proc); 1234 } 1235 } else /* if (pgid < 0) */ { 1236 sx_slock(&proctree_lock); 1237 SIGIO_LOCK(); 1238 osigio = funsetown_locked(*sigiop); 1239 pgrp = pgfind(-pgid); 1240 if (pgrp == NULL) { 1241 ret = ESRCH; 1242 } else { 1243 if (pgrp->pg_session != curthread->td_proc->p_session) { 1244 /* 1245 * Policy - Don't allow a process to FSETOWN a 1246 * process in another session. 1247 * 1248 * Remove this test to allow maximum flexibility 1249 * or restrict FSETOWN to the current process or 1250 * process group for maximum safety. 1251 */ 1252 ret = EPERM; 1253 } else { 1254 sigio->sio_pgrp = pgrp; 1255 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, 1256 sio_pgsigio); 1257 } 1258 PGRP_UNLOCK(pgrp); 1259 } 1260 sx_sunlock(&proctree_lock); 1261 } 1262 if (ret == 0) 1263 *sigiop = sigio; 1264 SIGIO_UNLOCK(); 1265 if (osigio != NULL) 1266 sigiofree(osigio); 1267 return (ret); 1268 } 1269 1270 /* 1271 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1272 */ 1273 pid_t 1274 fgetown(struct sigio **sigiop) 1275 { 1276 pid_t pgid; 1277 1278 SIGIO_LOCK(); 1279 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1280 SIGIO_UNLOCK(); 1281 return (pgid); 1282 } 1283 1284 static int 1285 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1286 bool audit) 1287 { 1288 int error; 1289 1290 FILEDESC_XLOCK_ASSERT(fdp); 1291 1292 /* 1293 * We now hold the fp reference that used to be owned by the 1294 * descriptor array. We have to unlock the FILEDESC *AFTER* 1295 * knote_fdclose to prevent a race of the fd getting opened, a knote 1296 * added, and deleteing a knote for the new fd. 1297 */ 1298 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) 1299 knote_fdclose(td, fd); 1300 1301 /* 1302 * We need to notify mqueue if the object is of type mqueue. 1303 */ 1304 if (__predict_false(fp->f_type == DTYPE_MQUEUE)) 1305 mq_fdclose(td, fd, fp); 1306 FILEDESC_XUNLOCK(fdp); 1307 1308 #ifdef AUDIT 1309 if (AUDITING_TD(td) && audit) 1310 audit_sysclose(td, fd, fp); 1311 #endif 1312 error = closef(fp, td); 1313 1314 /* 1315 * All paths leading up to closefp() will have already removed or 1316 * replaced the fd in the filedesc table, so a restart would not 1317 * operate on the same file. 1318 */ 1319 if (error == ERESTART) 1320 error = EINTR; 1321 1322 return (error); 1323 } 1324 1325 static int 1326 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1327 bool holdleaders, bool audit) 1328 { 1329 int error; 1330 1331 FILEDESC_XLOCK_ASSERT(fdp); 1332 1333 if (holdleaders) { 1334 if (td->td_proc->p_fdtol != NULL) { 1335 /* 1336 * Ask fdfree() to sleep to ensure that all relevant 1337 * process leaders can be traversed in closef(). 1338 */ 1339 fdp->fd_holdleaderscount++; 1340 } else { 1341 holdleaders = false; 1342 } 1343 } 1344 1345 error = closefp_impl(fdp, fd, fp, td, audit); 1346 if (holdleaders) { 1347 FILEDESC_XLOCK(fdp); 1348 fdp->fd_holdleaderscount--; 1349 if (fdp->fd_holdleaderscount == 0 && 1350 fdp->fd_holdleaderswakeup != 0) { 1351 fdp->fd_holdleaderswakeup = 0; 1352 wakeup(&fdp->fd_holdleaderscount); 1353 } 1354 FILEDESC_XUNLOCK(fdp); 1355 } 1356 return (error); 1357 } 1358 1359 static int 1360 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1361 bool holdleaders, bool audit) 1362 { 1363 1364 FILEDESC_XLOCK_ASSERT(fdp); 1365 1366 if (__predict_false(td->td_proc->p_fdtol != NULL)) { 1367 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit)); 1368 } else { 1369 return (closefp_impl(fdp, fd, fp, td, audit)); 1370 } 1371 } 1372 1373 /* 1374 * Close a file descriptor. 1375 */ 1376 #ifndef _SYS_SYSPROTO_H_ 1377 struct close_args { 1378 int fd; 1379 }; 1380 #endif 1381 /* ARGSUSED */ 1382 int 1383 sys_close(struct thread *td, struct close_args *uap) 1384 { 1385 1386 return (kern_close(td, uap->fd)); 1387 } 1388 1389 int 1390 kern_close(struct thread *td, int fd) 1391 { 1392 struct filedesc *fdp; 1393 struct file *fp; 1394 1395 fdp = td->td_proc->p_fd; 1396 1397 FILEDESC_XLOCK(fdp); 1398 if ((fp = fget_noref(fdp, fd)) == NULL) { 1399 FILEDESC_XUNLOCK(fdp); 1400 return (EBADF); 1401 } 1402 fdfree(fdp, fd); 1403 1404 /* closefp() drops the FILEDESC lock for us. */ 1405 return (closefp(fdp, fd, fp, td, true, true)); 1406 } 1407 1408 static int 1409 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd) 1410 { 1411 struct filedesc *fdp; 1412 struct fdescenttbl *fdt; 1413 struct filedescent *fde; 1414 int fd; 1415 1416 fdp = td->td_proc->p_fd; 1417 FILEDESC_XLOCK(fdp); 1418 fdt = atomic_load_ptr(&fdp->fd_files); 1419 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1420 fd = lowfd; 1421 if (__predict_false(fd > highfd)) { 1422 goto out_locked; 1423 } 1424 for (; fd <= highfd; fd++) { 1425 fde = &fdt->fdt_ofiles[fd]; 1426 if (fde->fde_file != NULL) 1427 fde->fde_flags |= UF_EXCLOSE; 1428 } 1429 out_locked: 1430 FILEDESC_XUNLOCK(fdp); 1431 return (0); 1432 } 1433 1434 static int 1435 close_range_impl(struct thread *td, u_int lowfd, u_int highfd) 1436 { 1437 struct filedesc *fdp; 1438 const struct fdescenttbl *fdt; 1439 struct file *fp; 1440 int fd; 1441 1442 fdp = td->td_proc->p_fd; 1443 FILEDESC_XLOCK(fdp); 1444 fdt = atomic_load_ptr(&fdp->fd_files); 1445 highfd = MIN(highfd, fdt->fdt_nfiles - 1); 1446 fd = lowfd; 1447 if (__predict_false(fd > highfd)) { 1448 goto out_locked; 1449 } 1450 for (;;) { 1451 fp = fdt->fdt_ofiles[fd].fde_file; 1452 if (fp == NULL) { 1453 if (fd == highfd) 1454 goto out_locked; 1455 } else { 1456 fdfree(fdp, fd); 1457 (void) closefp(fdp, fd, fp, td, true, true); 1458 if (fd == highfd) 1459 goto out_unlocked; 1460 FILEDESC_XLOCK(fdp); 1461 fdt = atomic_load_ptr(&fdp->fd_files); 1462 } 1463 fd++; 1464 } 1465 out_locked: 1466 FILEDESC_XUNLOCK(fdp); 1467 out_unlocked: 1468 return (0); 1469 } 1470 1471 int 1472 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd) 1473 { 1474 1475 /* 1476 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 1477 * open should not be a usage error. From a close_range() perspective, 1478 * close_range(3, ~0U, 0) in the same scenario should also likely not 1479 * be a usage error as all fd above 3 are in-fact already closed. 1480 */ 1481 if (highfd < lowfd) { 1482 return (EINVAL); 1483 } 1484 1485 if ((flags & CLOSE_RANGE_CLOEXEC) != 0) 1486 return (close_range_cloexec(td, lowfd, highfd)); 1487 1488 return (close_range_impl(td, lowfd, highfd)); 1489 } 1490 1491 #ifndef _SYS_SYSPROTO_H_ 1492 struct close_range_args { 1493 u_int lowfd; 1494 u_int highfd; 1495 int flags; 1496 }; 1497 #endif 1498 int 1499 sys_close_range(struct thread *td, struct close_range_args *uap) 1500 { 1501 1502 AUDIT_ARG_FD(uap->lowfd); 1503 AUDIT_ARG_CMD(uap->highfd); 1504 AUDIT_ARG_FFLAGS(uap->flags); 1505 1506 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0) 1507 return (EINVAL); 1508 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd)); 1509 } 1510 1511 #ifdef COMPAT_FREEBSD12 1512 /* 1513 * Close open file descriptors. 1514 */ 1515 #ifndef _SYS_SYSPROTO_H_ 1516 struct freebsd12_closefrom_args { 1517 int lowfd; 1518 }; 1519 #endif 1520 /* ARGSUSED */ 1521 int 1522 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) 1523 { 1524 u_int lowfd; 1525 1526 AUDIT_ARG_FD(uap->lowfd); 1527 1528 /* 1529 * Treat negative starting file descriptor values identical to 1530 * closefrom(0) which closes all files. 1531 */ 1532 lowfd = MAX(0, uap->lowfd); 1533 return (kern_close_range(td, 0, lowfd, ~0U)); 1534 } 1535 #endif /* COMPAT_FREEBSD12 */ 1536 1537 #if defined(COMPAT_43) 1538 /* 1539 * Return status information about a file descriptor. 1540 */ 1541 #ifndef _SYS_SYSPROTO_H_ 1542 struct ofstat_args { 1543 int fd; 1544 struct ostat *sb; 1545 }; 1546 #endif 1547 /* ARGSUSED */ 1548 int 1549 ofstat(struct thread *td, struct ofstat_args *uap) 1550 { 1551 struct ostat oub; 1552 struct stat ub; 1553 int error; 1554 1555 error = kern_fstat(td, uap->fd, &ub); 1556 if (error == 0) { 1557 cvtstat(&ub, &oub); 1558 error = copyout(&oub, uap->sb, sizeof(oub)); 1559 } 1560 return (error); 1561 } 1562 #endif /* COMPAT_43 */ 1563 1564 #if defined(COMPAT_FREEBSD11) 1565 int 1566 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) 1567 { 1568 struct stat sb; 1569 struct freebsd11_stat osb; 1570 int error; 1571 1572 error = kern_fstat(td, uap->fd, &sb); 1573 if (error != 0) 1574 return (error); 1575 error = freebsd11_cvtstat(&sb, &osb); 1576 if (error == 0) 1577 error = copyout(&osb, uap->sb, sizeof(osb)); 1578 return (error); 1579 } 1580 #endif /* COMPAT_FREEBSD11 */ 1581 1582 /* 1583 * Return status information about a file descriptor. 1584 */ 1585 #ifndef _SYS_SYSPROTO_H_ 1586 struct fstat_args { 1587 int fd; 1588 struct stat *sb; 1589 }; 1590 #endif 1591 /* ARGSUSED */ 1592 int 1593 sys_fstat(struct thread *td, struct fstat_args *uap) 1594 { 1595 struct stat ub; 1596 int error; 1597 1598 error = kern_fstat(td, uap->fd, &ub); 1599 if (error == 0) 1600 error = copyout(&ub, uap->sb, sizeof(ub)); 1601 return (error); 1602 } 1603 1604 int 1605 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1606 { 1607 struct file *fp; 1608 int error; 1609 1610 AUDIT_ARG_FD(fd); 1611 1612 error = fget(td, fd, &cap_fstat_rights, &fp); 1613 if (__predict_false(error != 0)) 1614 return (error); 1615 1616 AUDIT_ARG_FILE(td->td_proc, fp); 1617 1618 error = fo_stat(fp, sbp, td->td_ucred); 1619 fdrop(fp, td); 1620 #ifdef __STAT_TIME_T_EXT 1621 sbp->st_atim_ext = 0; 1622 sbp->st_mtim_ext = 0; 1623 sbp->st_ctim_ext = 0; 1624 sbp->st_btim_ext = 0; 1625 #endif 1626 #ifdef KTRACE 1627 if (KTRPOINT(td, KTR_STRUCT)) 1628 ktrstat_error(sbp, error); 1629 #endif 1630 return (error); 1631 } 1632 1633 #if defined(COMPAT_FREEBSD11) 1634 /* 1635 * Return status information about a file descriptor. 1636 */ 1637 #ifndef _SYS_SYSPROTO_H_ 1638 struct freebsd11_nfstat_args { 1639 int fd; 1640 struct nstat *sb; 1641 }; 1642 #endif 1643 /* ARGSUSED */ 1644 int 1645 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) 1646 { 1647 struct nstat nub; 1648 struct stat ub; 1649 int error; 1650 1651 error = kern_fstat(td, uap->fd, &ub); 1652 if (error != 0) 1653 return (error); 1654 error = freebsd11_cvtnstat(&ub, &nub); 1655 if (error != 0) 1656 error = copyout(&nub, uap->sb, sizeof(nub)); 1657 return (error); 1658 } 1659 #endif /* COMPAT_FREEBSD11 */ 1660 1661 /* 1662 * Return pathconf information about a file descriptor. 1663 */ 1664 #ifndef _SYS_SYSPROTO_H_ 1665 struct fpathconf_args { 1666 int fd; 1667 int name; 1668 }; 1669 #endif 1670 /* ARGSUSED */ 1671 int 1672 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1673 { 1674 long value; 1675 int error; 1676 1677 error = kern_fpathconf(td, uap->fd, uap->name, &value); 1678 if (error == 0) 1679 td->td_retval[0] = value; 1680 return (error); 1681 } 1682 1683 int 1684 kern_fpathconf(struct thread *td, int fd, int name, long *valuep) 1685 { 1686 struct file *fp; 1687 struct vnode *vp; 1688 int error; 1689 1690 error = fget(td, fd, &cap_fpathconf_rights, &fp); 1691 if (error != 0) 1692 return (error); 1693 1694 if (name == _PC_ASYNC_IO) { 1695 *valuep = _POSIX_ASYNCHRONOUS_IO; 1696 goto out; 1697 } 1698 vp = fp->f_vnode; 1699 if (vp != NULL) { 1700 vn_lock(vp, LK_SHARED | LK_RETRY); 1701 error = VOP_PATHCONF(vp, name, valuep); 1702 VOP_UNLOCK(vp); 1703 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1704 if (name != _PC_PIPE_BUF) { 1705 error = EINVAL; 1706 } else { 1707 *valuep = PIPE_BUF; 1708 error = 0; 1709 } 1710 } else { 1711 error = EOPNOTSUPP; 1712 } 1713 out: 1714 fdrop(fp, td); 1715 return (error); 1716 } 1717 1718 /* 1719 * Copy filecaps structure allocating memory for ioctls array if needed. 1720 * 1721 * The last parameter indicates whether the fdtable is locked. If it is not and 1722 * ioctls are encountered, copying fails and the caller must lock the table. 1723 * 1724 * Note that if the table was not locked, the caller has to check the relevant 1725 * sequence counter to determine whether the operation was successful. 1726 */ 1727 bool 1728 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) 1729 { 1730 size_t size; 1731 1732 if (src->fc_ioctls != NULL && !locked) 1733 return (false); 1734 memcpy(dst, src, sizeof(*src)); 1735 if (src->fc_ioctls == NULL) 1736 return (true); 1737 1738 KASSERT(src->fc_nioctls > 0, 1739 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1740 1741 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1742 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1743 memcpy(dst->fc_ioctls, src->fc_ioctls, size); 1744 return (true); 1745 } 1746 1747 static u_long * 1748 filecaps_copy_prep(const struct filecaps *src) 1749 { 1750 u_long *ioctls; 1751 size_t size; 1752 1753 if (__predict_true(src->fc_ioctls == NULL)) 1754 return (NULL); 1755 1756 KASSERT(src->fc_nioctls > 0, 1757 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1758 1759 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1760 ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1761 return (ioctls); 1762 } 1763 1764 static void 1765 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, 1766 u_long *ioctls) 1767 { 1768 size_t size; 1769 1770 *dst = *src; 1771 if (__predict_true(src->fc_ioctls == NULL)) { 1772 MPASS(ioctls == NULL); 1773 return; 1774 } 1775 1776 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1777 dst->fc_ioctls = ioctls; 1778 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1779 } 1780 1781 /* 1782 * Move filecaps structure to the new place and clear the old place. 1783 */ 1784 void 1785 filecaps_move(struct filecaps *src, struct filecaps *dst) 1786 { 1787 1788 *dst = *src; 1789 bzero(src, sizeof(*src)); 1790 } 1791 1792 /* 1793 * Fill the given filecaps structure with full rights. 1794 */ 1795 static void 1796 filecaps_fill(struct filecaps *fcaps) 1797 { 1798 1799 CAP_ALL(&fcaps->fc_rights); 1800 fcaps->fc_ioctls = NULL; 1801 fcaps->fc_nioctls = -1; 1802 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1803 } 1804 1805 /* 1806 * Free memory allocated within filecaps structure. 1807 */ 1808 void 1809 filecaps_free(struct filecaps *fcaps) 1810 { 1811 1812 free(fcaps->fc_ioctls, M_FILECAPS); 1813 bzero(fcaps, sizeof(*fcaps)); 1814 } 1815 1816 static u_long * 1817 filecaps_free_prep(struct filecaps *fcaps) 1818 { 1819 u_long *ioctls; 1820 1821 ioctls = fcaps->fc_ioctls; 1822 bzero(fcaps, sizeof(*fcaps)); 1823 return (ioctls); 1824 } 1825 1826 static void 1827 filecaps_free_finish(u_long *ioctls) 1828 { 1829 1830 free(ioctls, M_FILECAPS); 1831 } 1832 1833 /* 1834 * Validate the given filecaps structure. 1835 */ 1836 static void 1837 filecaps_validate(const struct filecaps *fcaps, const char *func) 1838 { 1839 1840 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1841 ("%s: invalid rights", func)); 1842 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1843 ("%s: invalid fcntls", func)); 1844 KASSERT(fcaps->fc_fcntls == 0 || 1845 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1846 ("%s: fcntls without CAP_FCNTL", func)); 1847 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1848 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1849 ("%s: invalid ioctls", func)); 1850 KASSERT(fcaps->fc_nioctls == 0 || 1851 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1852 ("%s: ioctls without CAP_IOCTL", func)); 1853 } 1854 1855 static void 1856 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1857 { 1858 int nfd1; 1859 1860 FILEDESC_XLOCK_ASSERT(fdp); 1861 1862 nfd1 = fdp->fd_nfiles * 2; 1863 if (nfd1 < nfd) 1864 nfd1 = nfd; 1865 fdgrowtable(fdp, nfd1); 1866 } 1867 1868 /* 1869 * Grow the file table to accommodate (at least) nfd descriptors. 1870 */ 1871 static void 1872 fdgrowtable(struct filedesc *fdp, int nfd) 1873 { 1874 struct filedesc0 *fdp0; 1875 struct freetable *ft; 1876 struct fdescenttbl *ntable; 1877 struct fdescenttbl *otable; 1878 int nnfiles, onfiles; 1879 NDSLOTTYPE *nmap, *omap; 1880 1881 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1882 1883 /* save old values */ 1884 onfiles = fdp->fd_nfiles; 1885 otable = fdp->fd_files; 1886 omap = fdp->fd_map; 1887 1888 /* compute the size of the new table */ 1889 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1890 if (nnfiles <= onfiles) 1891 /* the table is already large enough */ 1892 return; 1893 1894 /* 1895 * Allocate a new table. We need enough space for the number of 1896 * entries, file entries themselves and the struct freetable we will use 1897 * when we decommission the table and place it on the freelist. 1898 * We place the struct freetable in the middle so we don't have 1899 * to worry about padding. 1900 */ 1901 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + 1902 nnfiles * sizeof(ntable->fdt_ofiles[0]) + 1903 sizeof(struct freetable), 1904 M_FILEDESC, M_ZERO | M_WAITOK); 1905 /* copy the old data */ 1906 ntable->fdt_nfiles = nnfiles; 1907 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, 1908 onfiles * sizeof(ntable->fdt_ofiles[0])); 1909 1910 /* 1911 * Allocate a new map only if the old is not large enough. It will 1912 * grow at a slower rate than the table as it can map more 1913 * entries than the table can hold. 1914 */ 1915 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1916 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1917 M_ZERO | M_WAITOK); 1918 /* copy over the old data and update the pointer */ 1919 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1920 fdp->fd_map = nmap; 1921 } 1922 1923 /* 1924 * Make sure that ntable is correctly initialized before we replace 1925 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent 1926 * data. 1927 */ 1928 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); 1929 1930 /* 1931 * Free the old file table when not shared by other threads or processes. 1932 * The old file table is considered to be shared when either are true: 1933 * - The process has more than one thread. 1934 * - The file descriptor table has been shared via fdshare(). 1935 * 1936 * When shared, the old file table will be placed on a freelist 1937 * which will be processed when the struct filedesc is released. 1938 * 1939 * Note that if onfiles == NDFILE, we're dealing with the original 1940 * static allocation contained within (struct filedesc0 *)fdp, 1941 * which must not be freed. 1942 */ 1943 if (onfiles > NDFILE) { 1944 /* 1945 * Note we may be called here from fdinit while allocating a 1946 * table for a new process in which case ->p_fd points 1947 * elsewhere. 1948 */ 1949 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) { 1950 free(otable, M_FILEDESC); 1951 } else { 1952 ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; 1953 fdp0 = (struct filedesc0 *)fdp; 1954 ft->ft_table = otable; 1955 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1956 } 1957 } 1958 /* 1959 * The map does not have the same possibility of threads still 1960 * holding references to it. So always free it as long as it 1961 * does not reference the original static allocation. 1962 */ 1963 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1964 free(omap, M_FILEDESC); 1965 } 1966 1967 /* 1968 * Allocate a file descriptor for the process. 1969 */ 1970 int 1971 fdalloc(struct thread *td, int minfd, int *result) 1972 { 1973 struct proc *p = td->td_proc; 1974 struct filedesc *fdp = p->p_fd; 1975 int fd, maxfd, allocfd; 1976 #ifdef RACCT 1977 int error; 1978 #endif 1979 1980 FILEDESC_XLOCK_ASSERT(fdp); 1981 1982 if (fdp->fd_freefile > minfd) 1983 minfd = fdp->fd_freefile; 1984 1985 maxfd = getmaxfd(td); 1986 1987 /* 1988 * Search the bitmap for a free descriptor starting at minfd. 1989 * If none is found, grow the file table. 1990 */ 1991 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1992 if (__predict_false(fd >= maxfd)) 1993 return (EMFILE); 1994 if (__predict_false(fd >= fdp->fd_nfiles)) { 1995 allocfd = min(fd * 2, maxfd); 1996 #ifdef RACCT 1997 if (RACCT_ENABLED()) { 1998 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); 1999 if (error != 0) 2000 return (EMFILE); 2001 } 2002 #endif 2003 /* 2004 * fd is already equal to first free descriptor >= minfd, so 2005 * we only need to grow the table and we are done. 2006 */ 2007 fdgrowtable_exp(fdp, allocfd); 2008 } 2009 2010 /* 2011 * Perform some sanity checks, then mark the file descriptor as 2012 * used and return it to the caller. 2013 */ 2014 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 2015 ("invalid descriptor %d", fd)); 2016 KASSERT(!fdisused(fdp, fd), 2017 ("fd_first_free() returned non-free descriptor")); 2018 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 2019 ("file descriptor isn't free")); 2020 fdused(fdp, fd); 2021 *result = fd; 2022 return (0); 2023 } 2024 2025 /* 2026 * Allocate n file descriptors for the process. 2027 */ 2028 int 2029 fdallocn(struct thread *td, int minfd, int *fds, int n) 2030 { 2031 struct proc *p = td->td_proc; 2032 struct filedesc *fdp = p->p_fd; 2033 int i; 2034 2035 FILEDESC_XLOCK_ASSERT(fdp); 2036 2037 for (i = 0; i < n; i++) 2038 if (fdalloc(td, 0, &fds[i]) != 0) 2039 break; 2040 2041 if (i < n) { 2042 for (i--; i >= 0; i--) 2043 fdunused(fdp, fds[i]); 2044 return (EMFILE); 2045 } 2046 2047 return (0); 2048 } 2049 2050 /* 2051 * Create a new open file structure and allocate a file descriptor for the 2052 * process that refers to it. We add one reference to the file for the 2053 * descriptor table and one reference for resultfp. This is to prevent us 2054 * being preempted and the entry in the descriptor table closed after we 2055 * release the FILEDESC lock. 2056 */ 2057 int 2058 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, 2059 struct filecaps *fcaps) 2060 { 2061 struct file *fp; 2062 int error, fd; 2063 2064 MPASS(resultfp != NULL); 2065 MPASS(resultfd != NULL); 2066 2067 error = _falloc_noinstall(td, &fp, 2); 2068 if (__predict_false(error != 0)) { 2069 return (error); 2070 } 2071 2072 error = finstall_refed(td, fp, &fd, flags, fcaps); 2073 if (__predict_false(error != 0)) { 2074 falloc_abort(td, fp); 2075 return (error); 2076 } 2077 2078 *resultfp = fp; 2079 *resultfd = fd; 2080 2081 return (0); 2082 } 2083 2084 /* 2085 * Create a new open file structure without allocating a file descriptor. 2086 */ 2087 int 2088 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n) 2089 { 2090 struct file *fp; 2091 int maxuserfiles = maxfiles - (maxfiles / 20); 2092 int openfiles_new; 2093 static struct timeval lastfail; 2094 static int curfail; 2095 2096 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 2097 MPASS(n > 0); 2098 2099 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; 2100 if ((openfiles_new >= maxuserfiles && 2101 priv_check(td, PRIV_MAXFILES) != 0) || 2102 openfiles_new >= maxfiles) { 2103 atomic_subtract_int(&openfiles, 1); 2104 if (ppsratecheck(&lastfail, &curfail, 1)) { 2105 printf("kern.maxfiles limit exceeded by uid %i, (%s) " 2106 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); 2107 } 2108 return (ENFILE); 2109 } 2110 fp = uma_zalloc(file_zone, M_WAITOK); 2111 bzero(fp, sizeof(*fp)); 2112 refcount_init(&fp->f_count, n); 2113 fp->f_cred = crhold(td->td_ucred); 2114 fp->f_ops = &badfileops; 2115 *resultfp = fp; 2116 return (0); 2117 } 2118 2119 void 2120 falloc_abort(struct thread *td, struct file *fp) 2121 { 2122 2123 /* 2124 * For assertion purposes. 2125 */ 2126 refcount_init(&fp->f_count, 0); 2127 _fdrop(fp, td); 2128 } 2129 2130 /* 2131 * Install a file in a file descriptor table. 2132 */ 2133 void 2134 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, 2135 struct filecaps *fcaps) 2136 { 2137 struct filedescent *fde; 2138 2139 MPASS(fp != NULL); 2140 if (fcaps != NULL) 2141 filecaps_validate(fcaps, __func__); 2142 FILEDESC_XLOCK_ASSERT(fdp); 2143 2144 fde = &fdp->fd_ofiles[fd]; 2145 #ifdef CAPABILITIES 2146 seqc_write_begin(&fde->fde_seqc); 2147 #endif 2148 fde->fde_file = fp; 2149 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; 2150 if (fcaps != NULL) 2151 filecaps_move(fcaps, &fde->fde_caps); 2152 else 2153 filecaps_fill(&fde->fde_caps); 2154 #ifdef CAPABILITIES 2155 seqc_write_end(&fde->fde_seqc); 2156 #endif 2157 } 2158 2159 int 2160 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags, 2161 struct filecaps *fcaps) 2162 { 2163 struct filedesc *fdp = td->td_proc->p_fd; 2164 int error; 2165 2166 MPASS(fd != NULL); 2167 2168 FILEDESC_XLOCK(fdp); 2169 error = fdalloc(td, 0, fd); 2170 if (__predict_true(error == 0)) { 2171 _finstall(fdp, fp, *fd, flags, fcaps); 2172 } 2173 FILEDESC_XUNLOCK(fdp); 2174 return (error); 2175 } 2176 2177 int 2178 finstall(struct thread *td, struct file *fp, int *fd, int flags, 2179 struct filecaps *fcaps) 2180 { 2181 int error; 2182 2183 MPASS(fd != NULL); 2184 2185 if (!fhold(fp)) 2186 return (EBADF); 2187 error = finstall_refed(td, fp, fd, flags, fcaps); 2188 if (__predict_false(error != 0)) { 2189 fdrop(fp, td); 2190 } 2191 return (error); 2192 } 2193 2194 /* 2195 * Build a new filedesc structure from another. 2196 * 2197 * If fdp is not NULL, return with it shared locked. 2198 */ 2199 struct filedesc * 2200 fdinit(void) 2201 { 2202 struct filedesc0 *newfdp0; 2203 struct filedesc *newfdp; 2204 2205 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); 2206 newfdp = &newfdp0->fd_fd; 2207 2208 /* Create the file descriptor table. */ 2209 FILEDESC_LOCK_INIT(newfdp); 2210 refcount_init(&newfdp->fd_refcnt, 1); 2211 refcount_init(&newfdp->fd_holdcnt, 1); 2212 newfdp->fd_map = newfdp0->fd_dmap; 2213 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; 2214 newfdp->fd_files->fdt_nfiles = NDFILE; 2215 2216 return (newfdp); 2217 } 2218 2219 /* 2220 * Build a pwddesc structure from another. 2221 * Copy the current, root, and jail root vnode references. 2222 * 2223 * If pdp is not NULL, return with it shared locked. 2224 */ 2225 struct pwddesc * 2226 pdinit(struct pwddesc *pdp, bool keeplock) 2227 { 2228 struct pwddesc *newpdp; 2229 struct pwd *newpwd; 2230 2231 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO); 2232 2233 PWDDESC_LOCK_INIT(newpdp); 2234 refcount_init(&newpdp->pd_refcount, 1); 2235 newpdp->pd_cmask = CMASK; 2236 2237 if (pdp == NULL) { 2238 newpwd = pwd_alloc(); 2239 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2240 return (newpdp); 2241 } 2242 2243 PWDDESC_XLOCK(pdp); 2244 newpwd = pwd_hold_pwddesc(pdp); 2245 smr_serialized_store(&newpdp->pd_pwd, newpwd, true); 2246 if (!keeplock) 2247 PWDDESC_XUNLOCK(pdp); 2248 return (newpdp); 2249 } 2250 2251 /* 2252 * Hold either filedesc or pwddesc of the passed process. 2253 * 2254 * The process lock is used to synchronize against the target exiting and 2255 * freeing the data. 2256 * 2257 * Clearing can be ilustrated in 3 steps: 2258 * 1. set the pointer to NULL. Either routine can race against it, hence 2259 * atomic_load_ptr. 2260 * 2. observe the process lock as not taken. Until then fdhold/pdhold can 2261 * race to either still see the pointer or find NULL. It is still safe to 2262 * grab a reference as clearing is stalled. 2263 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are 2264 * guaranteed to see NULL, making it safe to finish clearing 2265 */ 2266 static struct filedesc * 2267 fdhold(struct proc *p) 2268 { 2269 struct filedesc *fdp; 2270 2271 PROC_LOCK_ASSERT(p, MA_OWNED); 2272 fdp = atomic_load_ptr(&p->p_fd); 2273 if (fdp != NULL) 2274 refcount_acquire(&fdp->fd_holdcnt); 2275 return (fdp); 2276 } 2277 2278 static struct pwddesc * 2279 pdhold(struct proc *p) 2280 { 2281 struct pwddesc *pdp; 2282 2283 PROC_LOCK_ASSERT(p, MA_OWNED); 2284 pdp = atomic_load_ptr(&p->p_pd); 2285 if (pdp != NULL) 2286 refcount_acquire(&pdp->pd_refcount); 2287 return (pdp); 2288 } 2289 2290 static void 2291 fddrop(struct filedesc *fdp) 2292 { 2293 2294 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2295 if (refcount_release(&fdp->fd_holdcnt) == 0) 2296 return; 2297 } 2298 2299 FILEDESC_LOCK_DESTROY(fdp); 2300 uma_zfree(filedesc0_zone, fdp); 2301 } 2302 2303 static void 2304 pddrop(struct pwddesc *pdp) 2305 { 2306 struct pwd *pwd; 2307 2308 if (refcount_release_if_not_last(&pdp->pd_refcount)) 2309 return; 2310 2311 PWDDESC_XLOCK(pdp); 2312 if (refcount_release(&pdp->pd_refcount) == 0) { 2313 PWDDESC_XUNLOCK(pdp); 2314 return; 2315 } 2316 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 2317 pwd_set(pdp, NULL); 2318 PWDDESC_XUNLOCK(pdp); 2319 pwd_drop(pwd); 2320 2321 PWDDESC_LOCK_DESTROY(pdp); 2322 free(pdp, M_PWDDESC); 2323 } 2324 2325 /* 2326 * Share a filedesc structure. 2327 */ 2328 struct filedesc * 2329 fdshare(struct filedesc *fdp) 2330 { 2331 2332 refcount_acquire(&fdp->fd_refcnt); 2333 return (fdp); 2334 } 2335 2336 /* 2337 * Share a pwddesc structure. 2338 */ 2339 struct pwddesc * 2340 pdshare(struct pwddesc *pdp) 2341 { 2342 refcount_acquire(&pdp->pd_refcount); 2343 return (pdp); 2344 } 2345 2346 /* 2347 * Unshare a filedesc structure, if necessary by making a copy 2348 */ 2349 void 2350 fdunshare(struct thread *td) 2351 { 2352 struct filedesc *tmp; 2353 struct proc *p = td->td_proc; 2354 2355 if (refcount_load(&p->p_fd->fd_refcnt) == 1) 2356 return; 2357 2358 tmp = fdcopy(p->p_fd); 2359 fdescfree(td); 2360 p->p_fd = tmp; 2361 } 2362 2363 /* 2364 * Unshare a pwddesc structure. 2365 */ 2366 void 2367 pdunshare(struct thread *td) 2368 { 2369 struct pwddesc *pdp; 2370 struct proc *p; 2371 2372 p = td->td_proc; 2373 /* Not shared. */ 2374 if (refcount_load(&p->p_pd->pd_refcount) == 1) 2375 return; 2376 2377 pdp = pdcopy(p->p_pd); 2378 pdescfree(td); 2379 p->p_pd = pdp; 2380 } 2381 2382 /* 2383 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 2384 * this is to ease callers, not catch errors. 2385 */ 2386 struct filedesc * 2387 fdcopy(struct filedesc *fdp) 2388 { 2389 struct filedesc *newfdp; 2390 struct filedescent *nfde, *ofde; 2391 int i, lastfile; 2392 2393 MPASS(fdp != NULL); 2394 2395 newfdp = fdinit(); 2396 FILEDESC_SLOCK(fdp); 2397 for (;;) { 2398 lastfile = fdlastfile(fdp); 2399 if (lastfile < newfdp->fd_nfiles) 2400 break; 2401 FILEDESC_SUNLOCK(fdp); 2402 fdgrowtable(newfdp, lastfile + 1); 2403 FILEDESC_SLOCK(fdp); 2404 } 2405 /* copy all passable descriptors (i.e. not kqueue) */ 2406 newfdp->fd_freefile = fdp->fd_freefile; 2407 FILEDESC_FOREACH_FDE(fdp, i, ofde) { 2408 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 || 2409 !fhold(ofde->fde_file)) { 2410 if (newfdp->fd_freefile == fdp->fd_freefile) 2411 newfdp->fd_freefile = i; 2412 continue; 2413 } 2414 nfde = &newfdp->fd_ofiles[i]; 2415 *nfde = *ofde; 2416 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); 2417 fdused_init(newfdp, i); 2418 } 2419 MPASS(newfdp->fd_freefile != -1); 2420 FILEDESC_SUNLOCK(fdp); 2421 return (newfdp); 2422 } 2423 2424 /* 2425 * Copy a pwddesc structure. 2426 */ 2427 struct pwddesc * 2428 pdcopy(struct pwddesc *pdp) 2429 { 2430 struct pwddesc *newpdp; 2431 2432 MPASS(pdp != NULL); 2433 2434 newpdp = pdinit(pdp, true); 2435 newpdp->pd_cmask = pdp->pd_cmask; 2436 PWDDESC_XUNLOCK(pdp); 2437 return (newpdp); 2438 } 2439 2440 /* 2441 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. 2442 * one of processes using it exits) and the table used to be shared. 2443 */ 2444 static void 2445 fdclearlocks(struct thread *td) 2446 { 2447 struct filedesc *fdp; 2448 struct filedesc_to_leader *fdtol; 2449 struct flock lf; 2450 struct file *fp; 2451 struct proc *p; 2452 struct vnode *vp; 2453 int i; 2454 2455 p = td->td_proc; 2456 fdp = p->p_fd; 2457 fdtol = p->p_fdtol; 2458 MPASS(fdtol != NULL); 2459 2460 FILEDESC_XLOCK(fdp); 2461 KASSERT(fdtol->fdl_refcount > 0, 2462 ("filedesc_to_refcount botch: fdl_refcount=%d", 2463 fdtol->fdl_refcount)); 2464 if (fdtol->fdl_refcount == 1 && 2465 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2466 FILEDESC_FOREACH_FP(fdp, i, fp) { 2467 if (fp->f_type != DTYPE_VNODE || 2468 !fhold(fp)) 2469 continue; 2470 FILEDESC_XUNLOCK(fdp); 2471 lf.l_whence = SEEK_SET; 2472 lf.l_start = 0; 2473 lf.l_len = 0; 2474 lf.l_type = F_UNLCK; 2475 vp = fp->f_vnode; 2476 (void) VOP_ADVLOCK(vp, 2477 (caddr_t)p->p_leader, F_UNLCK, 2478 &lf, F_POSIX); 2479 FILEDESC_XLOCK(fdp); 2480 fdrop(fp, td); 2481 } 2482 } 2483 retry: 2484 if (fdtol->fdl_refcount == 1) { 2485 if (fdp->fd_holdleaderscount > 0 && 2486 (p->p_leader->p_flag & P_ADVLOCK) != 0) { 2487 /* 2488 * close() or kern_dup() has cleared a reference 2489 * in a shared file descriptor table. 2490 */ 2491 fdp->fd_holdleaderswakeup = 1; 2492 sx_sleep(&fdp->fd_holdleaderscount, 2493 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 2494 goto retry; 2495 } 2496 if (fdtol->fdl_holdcount > 0) { 2497 /* 2498 * Ensure that fdtol->fdl_leader remains 2499 * valid in closef(). 2500 */ 2501 fdtol->fdl_wakeup = 1; 2502 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 2503 "fdlhold", 0); 2504 goto retry; 2505 } 2506 } 2507 fdtol->fdl_refcount--; 2508 if (fdtol->fdl_refcount == 0 && 2509 fdtol->fdl_holdcount == 0) { 2510 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2511 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2512 } else 2513 fdtol = NULL; 2514 p->p_fdtol = NULL; 2515 FILEDESC_XUNLOCK(fdp); 2516 if (fdtol != NULL) 2517 free(fdtol, M_FILEDESC_TO_LEADER); 2518 } 2519 2520 /* 2521 * Release a filedesc structure. 2522 */ 2523 static void 2524 fdescfree_fds(struct thread *td, struct filedesc *fdp) 2525 { 2526 struct filedesc0 *fdp0; 2527 struct freetable *ft, *tft; 2528 struct filedescent *fde; 2529 struct file *fp; 2530 int i; 2531 2532 KASSERT(refcount_load(&fdp->fd_refcnt) == 0, 2533 ("%s: fd table %p carries references", __func__, fdp)); 2534 2535 /* 2536 * Serialize with threads iterating over the table, if any. 2537 */ 2538 if (refcount_load(&fdp->fd_holdcnt) > 1) { 2539 FILEDESC_XLOCK(fdp); 2540 FILEDESC_XUNLOCK(fdp); 2541 } 2542 2543 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2544 fp = fde->fde_file; 2545 fdefree_last(fde); 2546 (void) closef(fp, td); 2547 } 2548 2549 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2550 free(fdp->fd_map, M_FILEDESC); 2551 if (fdp->fd_nfiles > NDFILE) 2552 free(fdp->fd_files, M_FILEDESC); 2553 2554 fdp0 = (struct filedesc0 *)fdp; 2555 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) 2556 free(ft->ft_table, M_FILEDESC); 2557 2558 fddrop(fdp); 2559 } 2560 2561 void 2562 fdescfree(struct thread *td) 2563 { 2564 struct proc *p; 2565 struct filedesc *fdp; 2566 2567 p = td->td_proc; 2568 fdp = p->p_fd; 2569 MPASS(fdp != NULL); 2570 2571 #ifdef RACCT 2572 if (RACCT_ENABLED()) 2573 racct_set_unlocked(p, RACCT_NOFILE, 0); 2574 #endif 2575 2576 if (p->p_fdtol != NULL) 2577 fdclearlocks(td); 2578 2579 /* 2580 * Check fdhold for an explanation. 2581 */ 2582 atomic_store_ptr(&p->p_fd, NULL); 2583 atomic_thread_fence_seq_cst(); 2584 PROC_WAIT_UNLOCKED(p); 2585 2586 if (refcount_release(&fdp->fd_refcnt) == 0) 2587 return; 2588 2589 fdescfree_fds(td, fdp); 2590 } 2591 2592 void 2593 pdescfree(struct thread *td) 2594 { 2595 struct proc *p; 2596 struct pwddesc *pdp; 2597 2598 p = td->td_proc; 2599 pdp = p->p_pd; 2600 MPASS(pdp != NULL); 2601 2602 /* 2603 * Check pdhold for an explanation. 2604 */ 2605 atomic_store_ptr(&p->p_pd, NULL); 2606 atomic_thread_fence_seq_cst(); 2607 PROC_WAIT_UNLOCKED(p); 2608 2609 pddrop(pdp); 2610 } 2611 2612 /* 2613 * For setugid programs, we don't want to people to use that setugidness 2614 * to generate error messages which write to a file which otherwise would 2615 * otherwise be off-limits to the process. We check for filesystems where 2616 * the vnode can change out from under us after execve (like [lin]procfs). 2617 * 2618 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is 2619 * sufficient. We also don't check for setugidness since we know we are. 2620 */ 2621 static bool 2622 is_unsafe(struct file *fp) 2623 { 2624 struct vnode *vp; 2625 2626 if (fp->f_type != DTYPE_VNODE) 2627 return (false); 2628 2629 vp = fp->f_vnode; 2630 return ((vp->v_vflag & VV_PROCDEP) != 0); 2631 } 2632 2633 /* 2634 * Make this setguid thing safe, if at all possible. 2635 */ 2636 void 2637 fdsetugidsafety(struct thread *td) 2638 { 2639 struct filedesc *fdp; 2640 struct file *fp; 2641 int i; 2642 2643 fdp = td->td_proc->p_fd; 2644 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2645 ("the fdtable should not be shared")); 2646 MPASS(fdp->fd_nfiles >= 3); 2647 for (i = 0; i <= 2; i++) { 2648 fp = fdp->fd_ofiles[i].fde_file; 2649 if (fp != NULL && is_unsafe(fp)) { 2650 FILEDESC_XLOCK(fdp); 2651 knote_fdclose(td, i); 2652 /* 2653 * NULL-out descriptor prior to close to avoid 2654 * a race while close blocks. 2655 */ 2656 fdfree(fdp, i); 2657 FILEDESC_XUNLOCK(fdp); 2658 (void) closef(fp, td); 2659 } 2660 } 2661 } 2662 2663 /* 2664 * If a specific file object occupies a specific file descriptor, close the 2665 * file descriptor entry and drop a reference on the file object. This is a 2666 * convenience function to handle a subsequent error in a function that calls 2667 * falloc() that handles the race that another thread might have closed the 2668 * file descriptor out from under the thread creating the file object. 2669 */ 2670 void 2671 fdclose(struct thread *td, struct file *fp, int idx) 2672 { 2673 struct filedesc *fdp = td->td_proc->p_fd; 2674 2675 FILEDESC_XLOCK(fdp); 2676 if (fdp->fd_ofiles[idx].fde_file == fp) { 2677 fdfree(fdp, idx); 2678 FILEDESC_XUNLOCK(fdp); 2679 fdrop(fp, td); 2680 } else 2681 FILEDESC_XUNLOCK(fdp); 2682 } 2683 2684 /* 2685 * Close any files on exec? 2686 */ 2687 void 2688 fdcloseexec(struct thread *td) 2689 { 2690 struct filedesc *fdp; 2691 struct filedescent *fde; 2692 struct file *fp; 2693 int i; 2694 2695 fdp = td->td_proc->p_fd; 2696 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2697 ("the fdtable should not be shared")); 2698 FILEDESC_FOREACH_FDE(fdp, i, fde) { 2699 fp = fde->fde_file; 2700 if (fp->f_type == DTYPE_MQUEUE || 2701 (fde->fde_flags & UF_EXCLOSE)) { 2702 FILEDESC_XLOCK(fdp); 2703 fdfree(fdp, i); 2704 (void) closefp(fdp, i, fp, td, false, false); 2705 FILEDESC_UNLOCK_ASSERT(fdp); 2706 } 2707 } 2708 } 2709 2710 /* 2711 * It is unsafe for set[ug]id processes to be started with file 2712 * descriptors 0..2 closed, as these descriptors are given implicit 2713 * significance in the Standard C library. fdcheckstd() will create a 2714 * descriptor referencing /dev/null for each of stdin, stdout, and 2715 * stderr that is not already open. 2716 */ 2717 int 2718 fdcheckstd(struct thread *td) 2719 { 2720 struct filedesc *fdp; 2721 register_t save; 2722 int i, error, devnull; 2723 2724 fdp = td->td_proc->p_fd; 2725 KASSERT(refcount_load(&fdp->fd_refcnt) == 1, 2726 ("the fdtable should not be shared")); 2727 MPASS(fdp->fd_nfiles >= 3); 2728 devnull = -1; 2729 for (i = 0; i <= 2; i++) { 2730 if (fdp->fd_ofiles[i].fde_file != NULL) 2731 continue; 2732 2733 save = td->td_retval[0]; 2734 if (devnull != -1) { 2735 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); 2736 } else { 2737 error = kern_openat(td, AT_FDCWD, "/dev/null", 2738 UIO_SYSSPACE, O_RDWR, 0); 2739 if (error == 0) { 2740 devnull = td->td_retval[0]; 2741 KASSERT(devnull == i, ("we didn't get our fd")); 2742 } 2743 } 2744 td->td_retval[0] = save; 2745 if (error != 0) 2746 return (error); 2747 } 2748 return (0); 2749 } 2750 2751 /* 2752 * Internal form of close. Decrement reference count on file structure. 2753 * Note: td may be NULL when closing a file that was being passed in a 2754 * message. 2755 */ 2756 int 2757 closef(struct file *fp, struct thread *td) 2758 { 2759 struct vnode *vp; 2760 struct flock lf; 2761 struct filedesc_to_leader *fdtol; 2762 struct filedesc *fdp; 2763 2764 MPASS(td != NULL); 2765 2766 /* 2767 * POSIX record locking dictates that any close releases ALL 2768 * locks owned by this process. This is handled by setting 2769 * a flag in the unlock to free ONLY locks obeying POSIX 2770 * semantics, and not to free BSD-style file locks. 2771 * If the descriptor was in a message, POSIX-style locks 2772 * aren't passed with the descriptor, and the thread pointer 2773 * will be NULL. Callers should be careful only to pass a 2774 * NULL thread pointer when there really is no owning 2775 * context that might have locks, or the locks will be 2776 * leaked. 2777 */ 2778 if (fp->f_type == DTYPE_VNODE) { 2779 vp = fp->f_vnode; 2780 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2781 lf.l_whence = SEEK_SET; 2782 lf.l_start = 0; 2783 lf.l_len = 0; 2784 lf.l_type = F_UNLCK; 2785 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2786 F_UNLCK, &lf, F_POSIX); 2787 } 2788 fdtol = td->td_proc->p_fdtol; 2789 if (fdtol != NULL) { 2790 /* 2791 * Handle special case where file descriptor table is 2792 * shared between multiple process leaders. 2793 */ 2794 fdp = td->td_proc->p_fd; 2795 FILEDESC_XLOCK(fdp); 2796 for (fdtol = fdtol->fdl_next; 2797 fdtol != td->td_proc->p_fdtol; 2798 fdtol = fdtol->fdl_next) { 2799 if ((fdtol->fdl_leader->p_flag & 2800 P_ADVLOCK) == 0) 2801 continue; 2802 fdtol->fdl_holdcount++; 2803 FILEDESC_XUNLOCK(fdp); 2804 lf.l_whence = SEEK_SET; 2805 lf.l_start = 0; 2806 lf.l_len = 0; 2807 lf.l_type = F_UNLCK; 2808 vp = fp->f_vnode; 2809 (void) VOP_ADVLOCK(vp, 2810 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2811 F_POSIX); 2812 FILEDESC_XLOCK(fdp); 2813 fdtol->fdl_holdcount--; 2814 if (fdtol->fdl_holdcount == 0 && 2815 fdtol->fdl_wakeup != 0) { 2816 fdtol->fdl_wakeup = 0; 2817 wakeup(fdtol); 2818 } 2819 } 2820 FILEDESC_XUNLOCK(fdp); 2821 } 2822 } 2823 return (fdrop_close(fp, td)); 2824 } 2825 2826 /* 2827 * Hack for file descriptor passing code. 2828 */ 2829 void 2830 closef_nothread(struct file *fp) 2831 { 2832 2833 fdrop(fp, NULL); 2834 } 2835 2836 /* 2837 * Initialize the file pointer with the specified properties. 2838 * 2839 * The ops are set with release semantics to be certain that the flags, type, 2840 * and data are visible when ops is. This is to prevent ops methods from being 2841 * called with bad data. 2842 */ 2843 void 2844 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2845 { 2846 fp->f_data = data; 2847 fp->f_flag = flag; 2848 fp->f_type = type; 2849 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2850 } 2851 2852 void 2853 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops) 2854 { 2855 fp->f_seqcount[UIO_READ] = 1; 2856 fp->f_seqcount[UIO_WRITE] = 1; 2857 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, 2858 data, ops); 2859 } 2860 2861 int 2862 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2863 struct file **fpp, struct filecaps *havecapsp) 2864 { 2865 struct filedescent *fde; 2866 int error; 2867 2868 FILEDESC_LOCK_ASSERT(fdp); 2869 2870 *fpp = NULL; 2871 fde = fdeget_noref(fdp, fd); 2872 if (fde == NULL) { 2873 error = EBADF; 2874 goto out; 2875 } 2876 2877 #ifdef CAPABILITIES 2878 error = cap_check(cap_rights_fde_inline(fde), needrightsp); 2879 if (error != 0) 2880 goto out; 2881 #endif 2882 2883 if (havecapsp != NULL) 2884 filecaps_copy(&fde->fde_caps, havecapsp, true); 2885 2886 *fpp = fde->fde_file; 2887 2888 error = 0; 2889 out: 2890 return (error); 2891 } 2892 2893 #ifdef CAPABILITIES 2894 int 2895 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2896 struct file **fpp, struct filecaps *havecapsp) 2897 { 2898 struct filedesc *fdp = td->td_proc->p_fd; 2899 int error; 2900 struct file *fp; 2901 seqc_t seq; 2902 2903 *fpp = NULL; 2904 for (;;) { 2905 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq); 2906 if (error != 0) 2907 return (error); 2908 2909 if (havecapsp != NULL) { 2910 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, 2911 havecapsp, false)) { 2912 fdrop(fp, td); 2913 goto get_locked; 2914 } 2915 } 2916 2917 if (!fd_modified(fdp, fd, seq)) 2918 break; 2919 fdrop(fp, td); 2920 } 2921 2922 *fpp = fp; 2923 return (0); 2924 2925 get_locked: 2926 FILEDESC_SLOCK(fdp); 2927 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp); 2928 if (error == 0 && !fhold(*fpp)) 2929 error = EBADF; 2930 FILEDESC_SUNLOCK(fdp); 2931 return (error); 2932 } 2933 #else 2934 int 2935 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, 2936 struct file **fpp, struct filecaps *havecapsp) 2937 { 2938 int error; 2939 error = fget_unlocked(td, fd, needrightsp, fpp); 2940 if (havecapsp != NULL && error == 0) 2941 filecaps_fill(havecapsp); 2942 2943 return (error); 2944 } 2945 #endif 2946 2947 #ifdef CAPABILITIES 2948 int 2949 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 2950 { 2951 const struct filedescent *fde; 2952 const struct fdescenttbl *fdt; 2953 struct filedesc *fdp; 2954 struct file *fp; 2955 struct vnode *vp; 2956 const cap_rights_t *haverights; 2957 cap_rights_t rights; 2958 seqc_t seq; 2959 2960 VFS_SMR_ASSERT_ENTERED(); 2961 2962 rights = *ndp->ni_rightsneeded; 2963 cap_rights_set_one(&rights, CAP_LOOKUP); 2964 2965 fdp = curproc->p_fd; 2966 fdt = fdp->fd_files; 2967 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 2968 return (EBADF); 2969 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 2970 fde = &fdt->fdt_ofiles[fd]; 2971 haverights = cap_rights_fde_inline(fde); 2972 fp = fde->fde_file; 2973 if (__predict_false(fp == NULL)) 2974 return (EAGAIN); 2975 if (__predict_false(cap_check_inline_transient(haverights, &rights))) 2976 return (EAGAIN); 2977 *fsearch = ((fp->f_flag & FSEARCH) != 0); 2978 vp = fp->f_vnode; 2979 if (__predict_false(vp == NULL)) { 2980 return (EAGAIN); 2981 } 2982 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { 2983 return (EAGAIN); 2984 } 2985 /* 2986 * Use an acquire barrier to force re-reading of fdt so it is 2987 * refreshed for verification. 2988 */ 2989 atomic_thread_fence_acq(); 2990 fdt = fdp->fd_files; 2991 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 2992 return (EAGAIN); 2993 /* 2994 * If file descriptor doesn't have all rights, 2995 * all lookups relative to it must also be 2996 * strictly relative. 2997 * 2998 * Not yet supported by fast path. 2999 */ 3000 CAP_ALL(&rights); 3001 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || 3002 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 3003 ndp->ni_filecaps.fc_nioctls != -1) { 3004 #ifdef notyet 3005 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 3006 #else 3007 return (EAGAIN); 3008 #endif 3009 } 3010 *vpp = vp; 3011 return (0); 3012 } 3013 #else 3014 int 3015 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) 3016 { 3017 const struct fdescenttbl *fdt; 3018 struct filedesc *fdp; 3019 struct file *fp; 3020 struct vnode *vp; 3021 3022 VFS_SMR_ASSERT_ENTERED(); 3023 3024 fdp = curproc->p_fd; 3025 fdt = fdp->fd_files; 3026 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3027 return (EBADF); 3028 fp = fdt->fdt_ofiles[fd].fde_file; 3029 if (__predict_false(fp == NULL)) 3030 return (EAGAIN); 3031 *fsearch = ((fp->f_flag & FSEARCH) != 0); 3032 vp = fp->f_vnode; 3033 if (__predict_false(vp == NULL || vp->v_type != VDIR)) { 3034 return (EAGAIN); 3035 } 3036 /* 3037 * Use an acquire barrier to force re-reading of fdt so it is 3038 * refreshed for verification. 3039 */ 3040 atomic_thread_fence_acq(); 3041 fdt = fdp->fd_files; 3042 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3043 return (EAGAIN); 3044 filecaps_fill(&ndp->ni_filecaps); 3045 *vpp = vp; 3046 return (0); 3047 } 3048 #endif 3049 3050 /* 3051 * Fetch the descriptor locklessly. 3052 * 3053 * We avoid fdrop() races by never raising a refcount above 0. To accomplish 3054 * this we have to use a cmpset loop rather than an atomic_add. The descriptor 3055 * must be re-verified once we acquire a reference to be certain that the 3056 * identity is still correct and we did not lose a race due to preemption. 3057 * 3058 * Force a reload of fdt when looping. Another thread could reallocate 3059 * the table before this fd was closed, so it is possible that there is 3060 * a stale fp pointer in cached version. 3061 */ 3062 #ifdef CAPABILITIES 3063 static int 3064 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp, 3065 struct file **fpp, seqc_t *seqp) 3066 { 3067 struct filedesc *fdp; 3068 const struct filedescent *fde; 3069 const struct fdescenttbl *fdt; 3070 struct file *fp; 3071 seqc_t seq; 3072 cap_rights_t haverights; 3073 int error; 3074 3075 fdp = td->td_proc->p_fd; 3076 fdt = fdp->fd_files; 3077 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3078 return (EBADF); 3079 3080 for (;;) { 3081 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3082 fde = &fdt->fdt_ofiles[fd]; 3083 haverights = *cap_rights_fde_inline(fde); 3084 fp = fde->fde_file; 3085 if (__predict_false(fp == NULL)) { 3086 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3087 return (EBADF); 3088 fdt = atomic_load_ptr(&fdp->fd_files); 3089 continue; 3090 } 3091 error = cap_check_inline(&haverights, needrightsp); 3092 if (__predict_false(error != 0)) { 3093 if (seqc_consistent(fd_seqc(fdt, fd), seq)) 3094 return (error); 3095 fdt = atomic_load_ptr(&fdp->fd_files); 3096 continue; 3097 } 3098 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3099 fdt = atomic_load_ptr(&fdp->fd_files); 3100 continue; 3101 } 3102 /* 3103 * Use an acquire barrier to force re-reading of fdt so it is 3104 * refreshed for verification. 3105 */ 3106 atomic_thread_fence_acq(); 3107 fdt = fdp->fd_files; 3108 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)) 3109 break; 3110 fdrop(fp, td); 3111 } 3112 *fpp = fp; 3113 if (seqp != NULL) { 3114 *seqp = seq; 3115 } 3116 return (0); 3117 } 3118 #else 3119 static int 3120 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp, 3121 struct file **fpp, seqc_t *seqp __unused) 3122 { 3123 struct filedesc *fdp; 3124 const struct fdescenttbl *fdt; 3125 struct file *fp; 3126 3127 fdp = td->td_proc->p_fd; 3128 fdt = fdp->fd_files; 3129 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) 3130 return (EBADF); 3131 3132 for (;;) { 3133 fp = fdt->fdt_ofiles[fd].fde_file; 3134 if (__predict_false(fp == NULL)) 3135 return (EBADF); 3136 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { 3137 fdt = atomic_load_ptr(&fdp->fd_files); 3138 continue; 3139 } 3140 /* 3141 * Use an acquire barrier to force re-reading of fdt so it is 3142 * refreshed for verification. 3143 */ 3144 atomic_thread_fence_acq(); 3145 fdt = fdp->fd_files; 3146 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file)) 3147 break; 3148 fdrop(fp, td); 3149 } 3150 *fpp = fp; 3151 return (0); 3152 } 3153 #endif 3154 3155 /* 3156 * See the comments in fget_unlocked_seq for an explanation of how this works. 3157 * 3158 * This is a simplified variant which bails out to the aforementioned routine 3159 * if anything goes wrong. In practice this only happens when userspace is 3160 * racing with itself. 3161 */ 3162 int 3163 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp, 3164 struct file **fpp) 3165 { 3166 struct filedesc *fdp; 3167 #ifdef CAPABILITIES 3168 const struct filedescent *fde; 3169 #endif 3170 const struct fdescenttbl *fdt; 3171 struct file *fp; 3172 #ifdef CAPABILITIES 3173 seqc_t seq; 3174 const cap_rights_t *haverights; 3175 #endif 3176 3177 fdp = td->td_proc->p_fd; 3178 fdt = fdp->fd_files; 3179 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { 3180 *fpp = NULL; 3181 return (EBADF); 3182 } 3183 #ifdef CAPABILITIES 3184 seq = seqc_read_notmodify(fd_seqc(fdt, fd)); 3185 fde = &fdt->fdt_ofiles[fd]; 3186 haverights = cap_rights_fde_inline(fde); 3187 fp = fde->fde_file; 3188 #else 3189 fp = fdt->fdt_ofiles[fd].fde_file; 3190 #endif 3191 if (__predict_false(fp == NULL)) 3192 goto out_fallback; 3193 #ifdef CAPABILITIES 3194 if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) 3195 goto out_fallback; 3196 #endif 3197 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) 3198 goto out_fallback; 3199 3200 /* 3201 * Use an acquire barrier to force re-reading of fdt so it is 3202 * refreshed for verification. 3203 */ 3204 atomic_thread_fence_acq(); 3205 fdt = fdp->fd_files; 3206 #ifdef CAPABILITIES 3207 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))) 3208 #else 3209 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) 3210 #endif 3211 goto out_fdrop; 3212 *fpp = fp; 3213 return (0); 3214 out_fdrop: 3215 fdrop(fp, td); 3216 out_fallback: 3217 *fpp = NULL; 3218 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL)); 3219 } 3220 3221 /* 3222 * Translate fd -> file when the caller guarantees the file descriptor table 3223 * can't be changed by others. 3224 * 3225 * Note this does not mean the file object itself is only visible to the caller, 3226 * merely that it wont disappear without having to be referenced. 3227 * 3228 * Must be paired with fput_only_user. 3229 */ 3230 #ifdef CAPABILITIES 3231 int 3232 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3233 struct file **fpp) 3234 { 3235 const struct filedescent *fde; 3236 const struct fdescenttbl *fdt; 3237 const cap_rights_t *haverights; 3238 struct file *fp; 3239 int error; 3240 3241 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3242 3243 *fpp = NULL; 3244 if (__predict_false(fd >= fdp->fd_nfiles)) 3245 return (EBADF); 3246 3247 fdt = fdp->fd_files; 3248 fde = &fdt->fdt_ofiles[fd]; 3249 fp = fde->fde_file; 3250 if (__predict_false(fp == NULL)) 3251 return (EBADF); 3252 MPASS(refcount_load(&fp->f_count) > 0); 3253 haverights = cap_rights_fde_inline(fde); 3254 error = cap_check_inline(haverights, needrightsp); 3255 if (__predict_false(error != 0)) 3256 return (error); 3257 *fpp = fp; 3258 return (0); 3259 } 3260 #else 3261 int 3262 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 3263 struct file **fpp) 3264 { 3265 struct file *fp; 3266 3267 MPASS(FILEDESC_IS_ONLY_USER(fdp)); 3268 3269 *fpp = NULL; 3270 if (__predict_false(fd >= fdp->fd_nfiles)) 3271 return (EBADF); 3272 3273 fp = fdp->fd_ofiles[fd].fde_file; 3274 if (__predict_false(fp == NULL)) 3275 return (EBADF); 3276 3277 MPASS(refcount_load(&fp->f_count) > 0); 3278 *fpp = fp; 3279 return (0); 3280 } 3281 #endif 3282 3283 /* 3284 * Extract the file pointer associated with the specified descriptor for the 3285 * current user process. 3286 * 3287 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 3288 * returned. 3289 * 3290 * File's rights will be checked against the capability rights mask. 3291 * 3292 * If an error occurred the non-zero error is returned and *fpp is set to 3293 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 3294 * responsible for fdrop(). 3295 */ 3296 static __inline int 3297 _fget(struct thread *td, int fd, struct file **fpp, int flags, 3298 cap_rights_t *needrightsp) 3299 { 3300 struct file *fp; 3301 int error; 3302 3303 *fpp = NULL; 3304 error = fget_unlocked(td, fd, needrightsp, &fp); 3305 if (__predict_false(error != 0)) 3306 return (error); 3307 if (__predict_false(fp->f_ops == &badfileops)) { 3308 fdrop(fp, td); 3309 return (EBADF); 3310 } 3311 3312 /* 3313 * FREAD and FWRITE failure return EBADF as per POSIX. 3314 */ 3315 error = 0; 3316 switch (flags) { 3317 case FREAD: 3318 case FWRITE: 3319 if ((fp->f_flag & flags) == 0) 3320 error = EBADF; 3321 break; 3322 case FEXEC: 3323 if (fp->f_ops != &path_fileops && 3324 ((fp->f_flag & (FREAD | FEXEC)) == 0 || 3325 (fp->f_flag & FWRITE) != 0)) 3326 error = EBADF; 3327 break; 3328 case 0: 3329 break; 3330 default: 3331 KASSERT(0, ("wrong flags")); 3332 } 3333 3334 if (error != 0) { 3335 fdrop(fp, td); 3336 return (error); 3337 } 3338 3339 *fpp = fp; 3340 return (0); 3341 } 3342 3343 int 3344 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3345 { 3346 3347 return (_fget(td, fd, fpp, 0, rightsp)); 3348 } 3349 3350 int 3351 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, 3352 struct file **fpp) 3353 { 3354 int error; 3355 #ifndef CAPABILITIES 3356 error = _fget(td, fd, fpp, 0, rightsp); 3357 if (maxprotp != NULL) 3358 *maxprotp = VM_PROT_ALL; 3359 return (error); 3360 #else 3361 cap_rights_t fdrights; 3362 struct filedesc *fdp; 3363 struct file *fp; 3364 seqc_t seq; 3365 3366 *fpp = NULL; 3367 fdp = td->td_proc->p_fd; 3368 MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); 3369 for (;;) { 3370 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq); 3371 if (__predict_false(error != 0)) 3372 return (error); 3373 if (__predict_false(fp->f_ops == &badfileops)) { 3374 fdrop(fp, td); 3375 return (EBADF); 3376 } 3377 if (maxprotp != NULL) 3378 fdrights = *cap_rights(fdp, fd); 3379 if (!fd_modified(fdp, fd, seq)) 3380 break; 3381 fdrop(fp, td); 3382 } 3383 3384 /* 3385 * If requested, convert capability rights to access flags. 3386 */ 3387 if (maxprotp != NULL) 3388 *maxprotp = cap_rights_to_vmprot(&fdrights); 3389 *fpp = fp; 3390 return (0); 3391 #endif 3392 } 3393 3394 int 3395 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3396 { 3397 3398 return (_fget(td, fd, fpp, FREAD, rightsp)); 3399 } 3400 3401 int 3402 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 3403 { 3404 3405 return (_fget(td, fd, fpp, FWRITE, rightsp)); 3406 } 3407 3408 int 3409 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, 3410 struct file **fpp) 3411 { 3412 #ifndef CAPABILITIES 3413 return (fget_unlocked(td, fd, rightsp, fpp)); 3414 #else 3415 struct filedesc *fdp = td->td_proc->p_fd; 3416 struct file *fp; 3417 int error; 3418 seqc_t seq; 3419 3420 *fpp = NULL; 3421 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); 3422 for (;;) { 3423 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq); 3424 if (error != 0) 3425 return (error); 3426 error = cap_fcntl_check(fdp, fd, needfcntl); 3427 if (!fd_modified(fdp, fd, seq)) 3428 break; 3429 fdrop(fp, td); 3430 } 3431 if (error != 0) { 3432 fdrop(fp, td); 3433 return (error); 3434 } 3435 *fpp = fp; 3436 return (0); 3437 #endif 3438 } 3439 3440 /* 3441 * Like fget() but loads the underlying vnode, or returns an error if the 3442 * descriptor does not represent a vnode. Note that pipes use vnodes but 3443 * never have VM objects. The returned vnode will be vref()'d. 3444 * 3445 * XXX: what about the unused flags ? 3446 */ 3447 static __inline int 3448 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 3449 struct vnode **vpp) 3450 { 3451 struct file *fp; 3452 int error; 3453 3454 *vpp = NULL; 3455 error = _fget(td, fd, &fp, flags, needrightsp); 3456 if (error != 0) 3457 return (error); 3458 if (fp->f_vnode == NULL) { 3459 error = EINVAL; 3460 } else { 3461 *vpp = fp->f_vnode; 3462 vref(*vpp); 3463 } 3464 fdrop(fp, td); 3465 3466 return (error); 3467 } 3468 3469 int 3470 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3471 { 3472 3473 return (_fgetvp(td, fd, 0, rightsp, vpp)); 3474 } 3475 3476 int 3477 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 3478 struct filecaps *havecaps, struct vnode **vpp) 3479 { 3480 struct filecaps caps; 3481 struct file *fp; 3482 int error; 3483 3484 error = fget_cap(td, fd, needrightsp, &fp, &caps); 3485 if (error != 0) 3486 return (error); 3487 if (fp->f_ops == &badfileops) { 3488 error = EBADF; 3489 goto out; 3490 } 3491 if (fp->f_vnode == NULL) { 3492 error = EINVAL; 3493 goto out; 3494 } 3495 3496 *havecaps = caps; 3497 *vpp = fp->f_vnode; 3498 vref(*vpp); 3499 fdrop(fp, td); 3500 3501 return (0); 3502 out: 3503 filecaps_free(&caps); 3504 fdrop(fp, td); 3505 return (error); 3506 } 3507 3508 int 3509 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3510 { 3511 3512 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 3513 } 3514 3515 int 3516 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 3517 { 3518 3519 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 3520 } 3521 3522 #ifdef notyet 3523 int 3524 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 3525 struct vnode **vpp) 3526 { 3527 3528 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 3529 } 3530 #endif 3531 3532 /* 3533 * Handle the last reference to a file being closed. 3534 * 3535 * Without the noinline attribute clang keeps inlining the func thorough this 3536 * file when fdrop is used. 3537 */ 3538 int __noinline 3539 _fdrop(struct file *fp, struct thread *td) 3540 { 3541 int error; 3542 #ifdef INVARIANTS 3543 int count; 3544 3545 count = refcount_load(&fp->f_count); 3546 if (count != 0) 3547 panic("fdrop: fp %p count %d", fp, count); 3548 #endif 3549 error = fo_close(fp, td); 3550 atomic_subtract_int(&openfiles, 1); 3551 crfree(fp->f_cred); 3552 free(fp->f_advice, M_FADVISE); 3553 uma_zfree(file_zone, fp); 3554 3555 return (error); 3556 } 3557 3558 /* 3559 * Apply an advisory lock on a file descriptor. 3560 * 3561 * Just attempt to get a record lock of the requested type on the entire file 3562 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 3563 */ 3564 #ifndef _SYS_SYSPROTO_H_ 3565 struct flock_args { 3566 int fd; 3567 int how; 3568 }; 3569 #endif 3570 /* ARGSUSED */ 3571 int 3572 sys_flock(struct thread *td, struct flock_args *uap) 3573 { 3574 struct file *fp; 3575 struct vnode *vp; 3576 struct flock lf; 3577 int error; 3578 3579 error = fget(td, uap->fd, &cap_flock_rights, &fp); 3580 if (error != 0) 3581 return (error); 3582 error = EOPNOTSUPP; 3583 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { 3584 goto done; 3585 } 3586 if (fp->f_ops == &path_fileops) { 3587 goto done; 3588 } 3589 3590 error = 0; 3591 vp = fp->f_vnode; 3592 lf.l_whence = SEEK_SET; 3593 lf.l_start = 0; 3594 lf.l_len = 0; 3595 if (uap->how & LOCK_UN) { 3596 lf.l_type = F_UNLCK; 3597 atomic_clear_int(&fp->f_flag, FHASLOCK); 3598 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 3599 goto done; 3600 } 3601 if (uap->how & LOCK_EX) 3602 lf.l_type = F_WRLCK; 3603 else if (uap->how & LOCK_SH) 3604 lf.l_type = F_RDLCK; 3605 else { 3606 error = EBADF; 3607 goto done; 3608 } 3609 atomic_set_int(&fp->f_flag, FHASLOCK); 3610 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 3611 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 3612 done: 3613 fdrop(fp, td); 3614 return (error); 3615 } 3616 /* 3617 * Duplicate the specified descriptor to a free descriptor. 3618 */ 3619 int 3620 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 3621 int openerror, int *indxp) 3622 { 3623 struct filedescent *newfde, *oldfde; 3624 struct file *fp; 3625 u_long *ioctls; 3626 int error, indx; 3627 3628 KASSERT(openerror == ENODEV || openerror == ENXIO, 3629 ("unexpected error %d in %s", openerror, __func__)); 3630 3631 /* 3632 * If the to-be-dup'd fd number is greater than the allowed number 3633 * of file descriptors, or the fd to be dup'd has already been 3634 * closed, then reject. 3635 */ 3636 FILEDESC_XLOCK(fdp); 3637 if ((fp = fget_noref(fdp, dfd)) == NULL) { 3638 FILEDESC_XUNLOCK(fdp); 3639 return (EBADF); 3640 } 3641 3642 error = fdalloc(td, 0, &indx); 3643 if (error != 0) { 3644 FILEDESC_XUNLOCK(fdp); 3645 return (error); 3646 } 3647 3648 /* 3649 * There are two cases of interest here. 3650 * 3651 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 3652 * 3653 * For ENXIO steal away the file structure from (dfd) and store it in 3654 * (indx). (dfd) is effectively closed by this operation. 3655 */ 3656 switch (openerror) { 3657 case ENODEV: 3658 /* 3659 * Check that the mode the file is being opened for is a 3660 * subset of the mode of the existing descriptor. 3661 */ 3662 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 3663 fdunused(fdp, indx); 3664 FILEDESC_XUNLOCK(fdp); 3665 return (EACCES); 3666 } 3667 if (!fhold(fp)) { 3668 fdunused(fdp, indx); 3669 FILEDESC_XUNLOCK(fdp); 3670 return (EBADF); 3671 } 3672 newfde = &fdp->fd_ofiles[indx]; 3673 oldfde = &fdp->fd_ofiles[dfd]; 3674 ioctls = filecaps_copy_prep(&oldfde->fde_caps); 3675 #ifdef CAPABILITIES 3676 seqc_write_begin(&newfde->fde_seqc); 3677 #endif 3678 fde_copy(oldfde, newfde); 3679 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, 3680 ioctls); 3681 #ifdef CAPABILITIES 3682 seqc_write_end(&newfde->fde_seqc); 3683 #endif 3684 break; 3685 case ENXIO: 3686 /* 3687 * Steal away the file pointer from dfd and stuff it into indx. 3688 */ 3689 newfde = &fdp->fd_ofiles[indx]; 3690 oldfde = &fdp->fd_ofiles[dfd]; 3691 #ifdef CAPABILITIES 3692 seqc_write_begin(&oldfde->fde_seqc); 3693 seqc_write_begin(&newfde->fde_seqc); 3694 #endif 3695 fde_copy(oldfde, newfde); 3696 oldfde->fde_file = NULL; 3697 fdunused(fdp, dfd); 3698 #ifdef CAPABILITIES 3699 seqc_write_end(&newfde->fde_seqc); 3700 seqc_write_end(&oldfde->fde_seqc); 3701 #endif 3702 break; 3703 } 3704 FILEDESC_XUNLOCK(fdp); 3705 *indxp = indx; 3706 return (0); 3707 } 3708 3709 /* 3710 * This sysctl determines if we will allow a process to chroot(2) if it 3711 * has a directory open: 3712 * 0: disallowed for all processes. 3713 * 1: allowed for processes that were not already chroot(2)'ed. 3714 * 2: allowed for all processes. 3715 */ 3716 3717 static int chroot_allow_open_directories = 1; 3718 3719 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 3720 &chroot_allow_open_directories, 0, 3721 "Allow a process to chroot(2) if it has a directory open"); 3722 3723 /* 3724 * Helper function for raised chroot(2) security function: Refuse if 3725 * any filedescriptors are open directories. 3726 */ 3727 static int 3728 chroot_refuse_vdir_fds(struct filedesc *fdp) 3729 { 3730 struct vnode *vp; 3731 struct file *fp; 3732 int i; 3733 3734 FILEDESC_LOCK_ASSERT(fdp); 3735 3736 FILEDESC_FOREACH_FP(fdp, i, fp) { 3737 if (fp->f_type == DTYPE_VNODE) { 3738 vp = fp->f_vnode; 3739 if (vp->v_type == VDIR) 3740 return (EPERM); 3741 } 3742 } 3743 return (0); 3744 } 3745 3746 static void 3747 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) 3748 { 3749 3750 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { 3751 vrefact(oldpwd->pwd_cdir); 3752 newpwd->pwd_cdir = oldpwd->pwd_cdir; 3753 } 3754 3755 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { 3756 vrefact(oldpwd->pwd_rdir); 3757 newpwd->pwd_rdir = oldpwd->pwd_rdir; 3758 } 3759 3760 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { 3761 vrefact(oldpwd->pwd_jdir); 3762 newpwd->pwd_jdir = oldpwd->pwd_jdir; 3763 } 3764 } 3765 3766 struct pwd * 3767 pwd_hold_pwddesc(struct pwddesc *pdp) 3768 { 3769 struct pwd *pwd; 3770 3771 PWDDESC_ASSERT_XLOCKED(pdp); 3772 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3773 if (pwd != NULL) 3774 refcount_acquire(&pwd->pwd_refcount); 3775 return (pwd); 3776 } 3777 3778 bool 3779 pwd_hold_smr(struct pwd *pwd) 3780 { 3781 3782 MPASS(pwd != NULL); 3783 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { 3784 return (true); 3785 } 3786 return (false); 3787 } 3788 3789 struct pwd * 3790 pwd_hold(struct thread *td) 3791 { 3792 struct pwddesc *pdp; 3793 struct pwd *pwd; 3794 3795 pdp = td->td_proc->p_pd; 3796 3797 vfs_smr_enter(); 3798 pwd = vfs_smr_entered_load(&pdp->pd_pwd); 3799 if (pwd_hold_smr(pwd)) { 3800 vfs_smr_exit(); 3801 return (pwd); 3802 } 3803 vfs_smr_exit(); 3804 PWDDESC_XLOCK(pdp); 3805 pwd = pwd_hold_pwddesc(pdp); 3806 MPASS(pwd != NULL); 3807 PWDDESC_XUNLOCK(pdp); 3808 return (pwd); 3809 } 3810 3811 struct pwd * 3812 pwd_hold_proc(struct proc *p) 3813 { 3814 struct pwddesc *pdp; 3815 struct pwd *pwd; 3816 3817 PROC_ASSERT_HELD(p); 3818 PROC_LOCK(p); 3819 pdp = pdhold(p); 3820 MPASS(pdp != NULL); 3821 PROC_UNLOCK(p); 3822 3823 PWDDESC_XLOCK(pdp); 3824 pwd = pwd_hold_pwddesc(pdp); 3825 MPASS(pwd != NULL); 3826 PWDDESC_XUNLOCK(pdp); 3827 pddrop(pdp); 3828 return (pwd); 3829 } 3830 3831 static struct pwd * 3832 pwd_alloc(void) 3833 { 3834 struct pwd *pwd; 3835 3836 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); 3837 bzero(pwd, sizeof(*pwd)); 3838 refcount_init(&pwd->pwd_refcount, 1); 3839 return (pwd); 3840 } 3841 3842 void 3843 pwd_drop(struct pwd *pwd) 3844 { 3845 3846 if (!refcount_release(&pwd->pwd_refcount)) 3847 return; 3848 3849 if (pwd->pwd_cdir != NULL) 3850 vrele(pwd->pwd_cdir); 3851 if (pwd->pwd_rdir != NULL) 3852 vrele(pwd->pwd_rdir); 3853 if (pwd->pwd_jdir != NULL) 3854 vrele(pwd->pwd_jdir); 3855 uma_zfree_smr(pwd_zone, pwd); 3856 } 3857 3858 /* 3859 * The caller is responsible for invoking priv_check() and 3860 * mac_vnode_check_chroot() to authorize this operation. 3861 */ 3862 int 3863 pwd_chroot(struct thread *td, struct vnode *vp) 3864 { 3865 struct pwddesc *pdp; 3866 struct filedesc *fdp; 3867 struct pwd *newpwd, *oldpwd; 3868 int error; 3869 3870 fdp = td->td_proc->p_fd; 3871 pdp = td->td_proc->p_pd; 3872 newpwd = pwd_alloc(); 3873 FILEDESC_SLOCK(fdp); 3874 PWDDESC_XLOCK(pdp); 3875 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3876 if (chroot_allow_open_directories == 0 || 3877 (chroot_allow_open_directories == 1 && 3878 oldpwd->pwd_rdir != rootvnode)) { 3879 error = chroot_refuse_vdir_fds(fdp); 3880 FILEDESC_SUNLOCK(fdp); 3881 if (error != 0) { 3882 PWDDESC_XUNLOCK(pdp); 3883 pwd_drop(newpwd); 3884 return (error); 3885 } 3886 } else { 3887 FILEDESC_SUNLOCK(fdp); 3888 } 3889 3890 vrefact(vp); 3891 newpwd->pwd_rdir = vp; 3892 if (oldpwd->pwd_jdir == NULL) { 3893 vrefact(vp); 3894 newpwd->pwd_jdir = vp; 3895 } 3896 pwd_fill(oldpwd, newpwd); 3897 pwd_set(pdp, newpwd); 3898 PWDDESC_XUNLOCK(pdp); 3899 pwd_drop(oldpwd); 3900 return (0); 3901 } 3902 3903 void 3904 pwd_chdir(struct thread *td, struct vnode *vp) 3905 { 3906 struct pwddesc *pdp; 3907 struct pwd *newpwd, *oldpwd; 3908 3909 VNPASS(vp->v_usecount > 0, vp); 3910 3911 newpwd = pwd_alloc(); 3912 pdp = td->td_proc->p_pd; 3913 PWDDESC_XLOCK(pdp); 3914 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3915 newpwd->pwd_cdir = vp; 3916 pwd_fill(oldpwd, newpwd); 3917 pwd_set(pdp, newpwd); 3918 PWDDESC_XUNLOCK(pdp); 3919 pwd_drop(oldpwd); 3920 } 3921 3922 /* 3923 * jail_attach(2) changes both root and working directories. 3924 */ 3925 int 3926 pwd_chroot_chdir(struct thread *td, struct vnode *vp) 3927 { 3928 struct pwddesc *pdp; 3929 struct filedesc *fdp; 3930 struct pwd *newpwd, *oldpwd; 3931 int error; 3932 3933 fdp = td->td_proc->p_fd; 3934 pdp = td->td_proc->p_pd; 3935 newpwd = pwd_alloc(); 3936 FILEDESC_SLOCK(fdp); 3937 PWDDESC_XLOCK(pdp); 3938 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3939 error = chroot_refuse_vdir_fds(fdp); 3940 FILEDESC_SUNLOCK(fdp); 3941 if (error != 0) { 3942 PWDDESC_XUNLOCK(pdp); 3943 pwd_drop(newpwd); 3944 return (error); 3945 } 3946 3947 vrefact(vp); 3948 newpwd->pwd_rdir = vp; 3949 vrefact(vp); 3950 newpwd->pwd_cdir = vp; 3951 if (oldpwd->pwd_jdir == NULL) { 3952 vrefact(vp); 3953 newpwd->pwd_jdir = vp; 3954 } 3955 pwd_fill(oldpwd, newpwd); 3956 pwd_set(pdp, newpwd); 3957 PWDDESC_XUNLOCK(pdp); 3958 pwd_drop(oldpwd); 3959 return (0); 3960 } 3961 3962 void 3963 pwd_ensure_dirs(void) 3964 { 3965 struct pwddesc *pdp; 3966 struct pwd *oldpwd, *newpwd; 3967 3968 pdp = curproc->p_pd; 3969 PWDDESC_XLOCK(pdp); 3970 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3971 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) { 3972 PWDDESC_XUNLOCK(pdp); 3973 return; 3974 } 3975 PWDDESC_XUNLOCK(pdp); 3976 3977 newpwd = pwd_alloc(); 3978 PWDDESC_XLOCK(pdp); 3979 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 3980 pwd_fill(oldpwd, newpwd); 3981 if (newpwd->pwd_cdir == NULL) { 3982 vrefact(rootvnode); 3983 newpwd->pwd_cdir = rootvnode; 3984 } 3985 if (newpwd->pwd_rdir == NULL) { 3986 vrefact(rootvnode); 3987 newpwd->pwd_rdir = rootvnode; 3988 } 3989 pwd_set(pdp, newpwd); 3990 PWDDESC_XUNLOCK(pdp); 3991 pwd_drop(oldpwd); 3992 } 3993 3994 void 3995 pwd_set_rootvnode(void) 3996 { 3997 struct pwddesc *pdp; 3998 struct pwd *oldpwd, *newpwd; 3999 4000 pdp = curproc->p_pd; 4001 4002 newpwd = pwd_alloc(); 4003 PWDDESC_XLOCK(pdp); 4004 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4005 vrefact(rootvnode); 4006 newpwd->pwd_cdir = rootvnode; 4007 vrefact(rootvnode); 4008 newpwd->pwd_rdir = rootvnode; 4009 pwd_fill(oldpwd, newpwd); 4010 pwd_set(pdp, newpwd); 4011 PWDDESC_XUNLOCK(pdp); 4012 pwd_drop(oldpwd); 4013 } 4014 4015 /* 4016 * Scan all active processes and prisons to see if any of them have a current 4017 * or root directory of `olddp'. If so, replace them with the new mount point. 4018 */ 4019 void 4020 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 4021 { 4022 struct pwddesc *pdp; 4023 struct pwd *newpwd, *oldpwd; 4024 struct prison *pr; 4025 struct proc *p; 4026 int nrele; 4027 4028 if (vrefcnt(olddp) == 1) 4029 return; 4030 nrele = 0; 4031 newpwd = pwd_alloc(); 4032 sx_slock(&allproc_lock); 4033 FOREACH_PROC_IN_SYSTEM(p) { 4034 PROC_LOCK(p); 4035 pdp = pdhold(p); 4036 PROC_UNLOCK(p); 4037 if (pdp == NULL) 4038 continue; 4039 PWDDESC_XLOCK(pdp); 4040 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4041 if (oldpwd == NULL || 4042 (oldpwd->pwd_cdir != olddp && 4043 oldpwd->pwd_rdir != olddp && 4044 oldpwd->pwd_jdir != olddp)) { 4045 PWDDESC_XUNLOCK(pdp); 4046 pddrop(pdp); 4047 continue; 4048 } 4049 if (oldpwd->pwd_cdir == olddp) { 4050 vrefact(newdp); 4051 newpwd->pwd_cdir = newdp; 4052 } 4053 if (oldpwd->pwd_rdir == olddp) { 4054 vrefact(newdp); 4055 newpwd->pwd_rdir = newdp; 4056 } 4057 if (oldpwd->pwd_jdir == olddp) { 4058 vrefact(newdp); 4059 newpwd->pwd_jdir = newdp; 4060 } 4061 pwd_fill(oldpwd, newpwd); 4062 pwd_set(pdp, newpwd); 4063 PWDDESC_XUNLOCK(pdp); 4064 pwd_drop(oldpwd); 4065 pddrop(pdp); 4066 newpwd = pwd_alloc(); 4067 } 4068 sx_sunlock(&allproc_lock); 4069 pwd_drop(newpwd); 4070 if (rootvnode == olddp) { 4071 vrefact(newdp); 4072 rootvnode = newdp; 4073 nrele++; 4074 } 4075 mtx_lock(&prison0.pr_mtx); 4076 if (prison0.pr_root == olddp) { 4077 vrefact(newdp); 4078 prison0.pr_root = newdp; 4079 nrele++; 4080 } 4081 mtx_unlock(&prison0.pr_mtx); 4082 sx_slock(&allprison_lock); 4083 TAILQ_FOREACH(pr, &allprison, pr_list) { 4084 mtx_lock(&pr->pr_mtx); 4085 if (pr->pr_root == olddp) { 4086 vrefact(newdp); 4087 pr->pr_root = newdp; 4088 nrele++; 4089 } 4090 mtx_unlock(&pr->pr_mtx); 4091 } 4092 sx_sunlock(&allprison_lock); 4093 while (nrele--) 4094 vrele(olddp); 4095 } 4096 4097 struct filedesc_to_leader * 4098 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 4099 { 4100 struct filedesc_to_leader *fdtol; 4101 4102 fdtol = malloc(sizeof(struct filedesc_to_leader), 4103 M_FILEDESC_TO_LEADER, M_WAITOK); 4104 fdtol->fdl_refcount = 1; 4105 fdtol->fdl_holdcount = 0; 4106 fdtol->fdl_wakeup = 0; 4107 fdtol->fdl_leader = leader; 4108 if (old != NULL) { 4109 FILEDESC_XLOCK(fdp); 4110 fdtol->fdl_next = old->fdl_next; 4111 fdtol->fdl_prev = old; 4112 old->fdl_next = fdtol; 4113 fdtol->fdl_next->fdl_prev = fdtol; 4114 FILEDESC_XUNLOCK(fdp); 4115 } else { 4116 fdtol->fdl_next = fdtol; 4117 fdtol->fdl_prev = fdtol; 4118 } 4119 return (fdtol); 4120 } 4121 4122 static int 4123 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) 4124 { 4125 NDSLOTTYPE *map; 4126 struct filedesc *fdp; 4127 u_int namelen; 4128 int count, off, minoff; 4129 4130 namelen = arg2; 4131 if (namelen != 1) 4132 return (EINVAL); 4133 4134 if (*(int *)arg1 != 0) 4135 return (EINVAL); 4136 4137 fdp = curproc->p_fd; 4138 count = 0; 4139 FILEDESC_SLOCK(fdp); 4140 map = fdp->fd_map; 4141 off = NDSLOT(fdp->fd_nfiles - 1); 4142 for (minoff = NDSLOT(0); off >= minoff; --off) 4143 count += bitcountl(map[off]); 4144 FILEDESC_SUNLOCK(fdp); 4145 4146 return (SYSCTL_OUT(req, &count, sizeof(count))); 4147 } 4148 4149 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, 4150 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, 4151 "Number of open file descriptors"); 4152 4153 /* 4154 * Get file structures globally. 4155 */ 4156 static int 4157 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 4158 { 4159 struct xfile xf; 4160 struct filedesc *fdp; 4161 struct file *fp; 4162 struct proc *p; 4163 int error, n; 4164 4165 error = sysctl_wire_old_buffer(req, 0); 4166 if (error != 0) 4167 return (error); 4168 if (req->oldptr == NULL) { 4169 n = 0; 4170 sx_slock(&allproc_lock); 4171 FOREACH_PROC_IN_SYSTEM(p) { 4172 PROC_LOCK(p); 4173 if (p->p_state == PRS_NEW) { 4174 PROC_UNLOCK(p); 4175 continue; 4176 } 4177 fdp = fdhold(p); 4178 PROC_UNLOCK(p); 4179 if (fdp == NULL) 4180 continue; 4181 /* overestimates sparse tables. */ 4182 n += fdp->fd_nfiles; 4183 fddrop(fdp); 4184 } 4185 sx_sunlock(&allproc_lock); 4186 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 4187 } 4188 error = 0; 4189 bzero(&xf, sizeof(xf)); 4190 xf.xf_size = sizeof(xf); 4191 sx_slock(&allproc_lock); 4192 FOREACH_PROC_IN_SYSTEM(p) { 4193 PROC_LOCK(p); 4194 if (p->p_state == PRS_NEW) { 4195 PROC_UNLOCK(p); 4196 continue; 4197 } 4198 if (p_cansee(req->td, p) != 0) { 4199 PROC_UNLOCK(p); 4200 continue; 4201 } 4202 xf.xf_pid = p->p_pid; 4203 xf.xf_uid = p->p_ucred->cr_uid; 4204 fdp = fdhold(p); 4205 PROC_UNLOCK(p); 4206 if (fdp == NULL) 4207 continue; 4208 FILEDESC_SLOCK(fdp); 4209 if (refcount_load(&fdp->fd_refcnt) == 0) 4210 goto nextproc; 4211 FILEDESC_FOREACH_FP(fdp, n, fp) { 4212 xf.xf_fd = n; 4213 xf.xf_file = (uintptr_t)fp; 4214 xf.xf_data = (uintptr_t)fp->f_data; 4215 xf.xf_vnode = (uintptr_t)fp->f_vnode; 4216 xf.xf_type = (uintptr_t)fp->f_type; 4217 xf.xf_count = refcount_load(&fp->f_count); 4218 xf.xf_msgcount = 0; 4219 xf.xf_offset = foffset_get(fp); 4220 xf.xf_flag = fp->f_flag; 4221 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 4222 4223 /* 4224 * There is no need to re-check the fdtable refcount 4225 * here since the filedesc lock is not dropped in the 4226 * loop body. 4227 */ 4228 if (error != 0) 4229 break; 4230 } 4231 nextproc: 4232 FILEDESC_SUNLOCK(fdp); 4233 fddrop(fdp); 4234 if (error) 4235 break; 4236 } 4237 sx_sunlock(&allproc_lock); 4238 return (error); 4239 } 4240 4241 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 4242 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 4243 4244 #ifdef KINFO_FILE_SIZE 4245 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 4246 #endif 4247 4248 static int 4249 xlate_fflags(int fflags) 4250 { 4251 static const struct { 4252 int fflag; 4253 int kf_fflag; 4254 } fflags_table[] = { 4255 { FAPPEND, KF_FLAG_APPEND }, 4256 { FASYNC, KF_FLAG_ASYNC }, 4257 { FFSYNC, KF_FLAG_FSYNC }, 4258 { FHASLOCK, KF_FLAG_HASLOCK }, 4259 { FNONBLOCK, KF_FLAG_NONBLOCK }, 4260 { FREAD, KF_FLAG_READ }, 4261 { FWRITE, KF_FLAG_WRITE }, 4262 { O_CREAT, KF_FLAG_CREAT }, 4263 { O_DIRECT, KF_FLAG_DIRECT }, 4264 { O_EXCL, KF_FLAG_EXCL }, 4265 { O_EXEC, KF_FLAG_EXEC }, 4266 { O_EXLOCK, KF_FLAG_EXLOCK }, 4267 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 4268 { O_SHLOCK, KF_FLAG_SHLOCK }, 4269 { O_TRUNC, KF_FLAG_TRUNC } 4270 }; 4271 unsigned int i; 4272 int kflags; 4273 4274 kflags = 0; 4275 for (i = 0; i < nitems(fflags_table); i++) 4276 if (fflags & fflags_table[i].fflag) 4277 kflags |= fflags_table[i].kf_fflag; 4278 return (kflags); 4279 } 4280 4281 /* Trim unused data from kf_path by truncating the structure size. */ 4282 void 4283 pack_kinfo(struct kinfo_file *kif) 4284 { 4285 4286 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 4287 strlen(kif->kf_path) + 1; 4288 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 4289 } 4290 4291 static void 4292 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, 4293 struct kinfo_file *kif, struct filedesc *fdp, int flags) 4294 { 4295 int error; 4296 4297 bzero(kif, sizeof(*kif)); 4298 4299 /* Set a default type to allow for empty fill_kinfo() methods. */ 4300 kif->kf_type = KF_TYPE_UNKNOWN; 4301 kif->kf_flags = xlate_fflags(fp->f_flag); 4302 if (rightsp != NULL) 4303 kif->kf_cap_rights = *rightsp; 4304 else 4305 cap_rights_init_zero(&kif->kf_cap_rights); 4306 kif->kf_fd = fd; 4307 kif->kf_ref_count = refcount_load(&fp->f_count); 4308 kif->kf_offset = foffset_get(fp); 4309 4310 /* 4311 * This may drop the filedesc lock, so the 'fp' cannot be 4312 * accessed after this call. 4313 */ 4314 error = fo_fill_kinfo(fp, kif, fdp); 4315 if (error == 0) 4316 kif->kf_status |= KF_ATTR_VALID; 4317 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4318 pack_kinfo(kif); 4319 else 4320 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4321 } 4322 4323 static void 4324 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, 4325 struct kinfo_file *kif, int flags) 4326 { 4327 int error; 4328 4329 bzero(kif, sizeof(*kif)); 4330 4331 kif->kf_type = KF_TYPE_VNODE; 4332 error = vn_fill_kinfo_vnode(vp, kif); 4333 if (error == 0) 4334 kif->kf_status |= KF_ATTR_VALID; 4335 kif->kf_flags = xlate_fflags(fflags); 4336 cap_rights_init_zero(&kif->kf_cap_rights); 4337 kif->kf_fd = fd; 4338 kif->kf_ref_count = -1; 4339 kif->kf_offset = -1; 4340 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) 4341 pack_kinfo(kif); 4342 else 4343 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); 4344 vrele(vp); 4345 } 4346 4347 struct export_fd_buf { 4348 struct filedesc *fdp; 4349 struct pwddesc *pdp; 4350 struct sbuf *sb; 4351 ssize_t remainder; 4352 struct kinfo_file kif; 4353 int flags; 4354 }; 4355 4356 static int 4357 export_kinfo_to_sb(struct export_fd_buf *efbuf) 4358 { 4359 struct kinfo_file *kif; 4360 4361 kif = &efbuf->kif; 4362 if (efbuf->remainder != -1) { 4363 if (efbuf->remainder < kif->kf_structsize) 4364 return (ENOMEM); 4365 efbuf->remainder -= kif->kf_structsize; 4366 } 4367 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0) 4368 return (sbuf_error(efbuf->sb)); 4369 return (0); 4370 } 4371 4372 static int 4373 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, 4374 struct export_fd_buf *efbuf) 4375 { 4376 int error; 4377 4378 if (efbuf->remainder == 0) 4379 return (ENOMEM); 4380 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, 4381 efbuf->flags); 4382 FILEDESC_SUNLOCK(efbuf->fdp); 4383 error = export_kinfo_to_sb(efbuf); 4384 FILEDESC_SLOCK(efbuf->fdp); 4385 return (error); 4386 } 4387 4388 static int 4389 export_vnode_to_sb(struct vnode *vp, int fd, int fflags, 4390 struct export_fd_buf *efbuf) 4391 { 4392 int error; 4393 4394 if (efbuf->remainder == 0) 4395 return (ENOMEM); 4396 if (efbuf->pdp != NULL) 4397 PWDDESC_XUNLOCK(efbuf->pdp); 4398 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); 4399 error = export_kinfo_to_sb(efbuf); 4400 if (efbuf->pdp != NULL) 4401 PWDDESC_XLOCK(efbuf->pdp); 4402 return (error); 4403 } 4404 4405 /* 4406 * Store a process file descriptor information to sbuf. 4407 * 4408 * Takes a locked proc as argument, and returns with the proc unlocked. 4409 */ 4410 int 4411 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, 4412 int flags) 4413 { 4414 struct file *fp; 4415 struct filedesc *fdp; 4416 struct pwddesc *pdp; 4417 struct export_fd_buf *efbuf; 4418 struct vnode *cttyvp, *textvp, *tracevp; 4419 struct pwd *pwd; 4420 int error, i; 4421 cap_rights_t rights; 4422 4423 PROC_LOCK_ASSERT(p, MA_OWNED); 4424 4425 /* ktrace vnode */ 4426 tracevp = ktr_get_tracevp(p, true); 4427 /* text vnode */ 4428 textvp = p->p_textvp; 4429 if (textvp != NULL) 4430 vrefact(textvp); 4431 /* Controlling tty. */ 4432 cttyvp = NULL; 4433 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 4434 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 4435 if (cttyvp != NULL) 4436 vrefact(cttyvp); 4437 } 4438 fdp = fdhold(p); 4439 pdp = pdhold(p); 4440 PROC_UNLOCK(p); 4441 4442 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4443 efbuf->fdp = NULL; 4444 efbuf->pdp = NULL; 4445 efbuf->sb = sb; 4446 efbuf->remainder = maxlen; 4447 efbuf->flags = flags; 4448 4449 error = 0; 4450 if (tracevp != NULL) 4451 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, 4452 FREAD | FWRITE, efbuf); 4453 if (error == 0 && textvp != NULL) 4454 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, 4455 efbuf); 4456 if (error == 0 && cttyvp != NULL) 4457 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, 4458 FREAD | FWRITE, efbuf); 4459 if (error != 0 || pdp == NULL || fdp == NULL) 4460 goto fail; 4461 efbuf->fdp = fdp; 4462 efbuf->pdp = pdp; 4463 PWDDESC_XLOCK(pdp); 4464 pwd = pwd_hold_pwddesc(pdp); 4465 if (pwd != NULL) { 4466 /* working directory */ 4467 if (pwd->pwd_cdir != NULL) { 4468 vrefact(pwd->pwd_cdir); 4469 error = export_vnode_to_sb(pwd->pwd_cdir, 4470 KF_FD_TYPE_CWD, FREAD, efbuf); 4471 } 4472 /* root directory */ 4473 if (error == 0 && pwd->pwd_rdir != NULL) { 4474 vrefact(pwd->pwd_rdir); 4475 error = export_vnode_to_sb(pwd->pwd_rdir, 4476 KF_FD_TYPE_ROOT, FREAD, efbuf); 4477 } 4478 /* jail directory */ 4479 if (error == 0 && pwd->pwd_jdir != NULL) { 4480 vrefact(pwd->pwd_jdir); 4481 error = export_vnode_to_sb(pwd->pwd_jdir, 4482 KF_FD_TYPE_JAIL, FREAD, efbuf); 4483 } 4484 } 4485 PWDDESC_XUNLOCK(pdp); 4486 if (error != 0) 4487 goto fail; 4488 if (pwd != NULL) 4489 pwd_drop(pwd); 4490 FILEDESC_SLOCK(fdp); 4491 if (refcount_load(&fdp->fd_refcnt) == 0) 4492 goto skip; 4493 FILEDESC_FOREACH_FP(fdp, i, fp) { 4494 #ifdef CAPABILITIES 4495 rights = *cap_rights(fdp, i); 4496 #else /* !CAPABILITIES */ 4497 rights = cap_no_rights; 4498 #endif 4499 /* 4500 * Create sysctl entry. It is OK to drop the filedesc 4501 * lock inside of export_file_to_sb() as we will 4502 * re-validate and re-evaluate its properties when the 4503 * loop continues. 4504 */ 4505 error = export_file_to_sb(fp, i, &rights, efbuf); 4506 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4507 break; 4508 } 4509 skip: 4510 FILEDESC_SUNLOCK(fdp); 4511 fail: 4512 if (fdp != NULL) 4513 fddrop(fdp); 4514 if (pdp != NULL) 4515 pddrop(pdp); 4516 free(efbuf, M_TEMP); 4517 return (error); 4518 } 4519 4520 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 4521 4522 /* 4523 * Get per-process file descriptors for use by procstat(1), et al. 4524 */ 4525 static int 4526 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 4527 { 4528 struct sbuf sb; 4529 struct proc *p; 4530 ssize_t maxlen; 4531 u_int namelen; 4532 int error, error2, *name; 4533 4534 namelen = arg2; 4535 if (namelen != 1) 4536 return (EINVAL); 4537 4538 name = (int *)arg1; 4539 4540 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 4541 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4542 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4543 if (error != 0) { 4544 sbuf_delete(&sb); 4545 return (error); 4546 } 4547 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4548 error = kern_proc_filedesc_out(p, &sb, maxlen, 4549 KERN_FILEDESC_PACK_KINFO); 4550 error2 = sbuf_finish(&sb); 4551 sbuf_delete(&sb); 4552 return (error != 0 ? error : error2); 4553 } 4554 4555 #ifdef COMPAT_FREEBSD7 4556 #ifdef KINFO_OFILE_SIZE 4557 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 4558 #endif 4559 4560 static void 4561 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) 4562 { 4563 4564 okif->kf_structsize = sizeof(*okif); 4565 okif->kf_type = kif->kf_type; 4566 okif->kf_fd = kif->kf_fd; 4567 okif->kf_ref_count = kif->kf_ref_count; 4568 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | 4569 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | 4570 KF_FLAG_DIRECT | KF_FLAG_HASLOCK); 4571 okif->kf_offset = kif->kf_offset; 4572 if (kif->kf_type == KF_TYPE_VNODE) 4573 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; 4574 else 4575 okif->kf_vnode_type = KF_VTYPE_VNON; 4576 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); 4577 if (kif->kf_type == KF_TYPE_SOCKET) { 4578 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; 4579 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; 4580 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; 4581 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; 4582 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; 4583 } else { 4584 okif->kf_sa_local.ss_family = AF_UNSPEC; 4585 okif->kf_sa_peer.ss_family = AF_UNSPEC; 4586 } 4587 } 4588 4589 static int 4590 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, 4591 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req) 4592 { 4593 int error; 4594 4595 vrefact(vp); 4596 PWDDESC_XUNLOCK(pdp); 4597 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); 4598 kinfo_to_okinfo(kif, okif); 4599 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4600 PWDDESC_XLOCK(pdp); 4601 return (error); 4602 } 4603 4604 /* 4605 * Get per-process file descriptors for use by procstat(1), et al. 4606 */ 4607 static int 4608 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 4609 { 4610 struct kinfo_ofile *okif; 4611 struct kinfo_file *kif; 4612 struct filedesc *fdp; 4613 struct pwddesc *pdp; 4614 struct pwd *pwd; 4615 u_int namelen; 4616 int error, i, *name; 4617 struct file *fp; 4618 struct proc *p; 4619 4620 namelen = arg2; 4621 if (namelen != 1) 4622 return (EINVAL); 4623 4624 name = (int *)arg1; 4625 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4626 if (error != 0) 4627 return (error); 4628 fdp = fdhold(p); 4629 if (fdp != NULL) 4630 pdp = pdhold(p); 4631 PROC_UNLOCK(p); 4632 if (fdp == NULL || pdp == NULL) { 4633 if (fdp != NULL) 4634 fddrop(fdp); 4635 return (ENOENT); 4636 } 4637 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 4638 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); 4639 PWDDESC_XLOCK(pdp); 4640 pwd = pwd_hold_pwddesc(pdp); 4641 if (pwd != NULL) { 4642 if (pwd->pwd_cdir != NULL) 4643 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, 4644 okif, pdp, req); 4645 if (pwd->pwd_rdir != NULL) 4646 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, 4647 okif, pdp, req); 4648 if (pwd->pwd_jdir != NULL) 4649 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, 4650 okif, pdp, req); 4651 } 4652 PWDDESC_XUNLOCK(pdp); 4653 if (pwd != NULL) 4654 pwd_drop(pwd); 4655 FILEDESC_SLOCK(fdp); 4656 if (refcount_load(&fdp->fd_refcnt) == 0) 4657 goto skip; 4658 FILEDESC_FOREACH_FP(fdp, i, fp) { 4659 export_file_to_kinfo(fp, i, NULL, kif, fdp, 4660 KERN_FILEDESC_PACK_KINFO); 4661 FILEDESC_SUNLOCK(fdp); 4662 kinfo_to_okinfo(kif, okif); 4663 error = SYSCTL_OUT(req, okif, sizeof(*okif)); 4664 FILEDESC_SLOCK(fdp); 4665 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0) 4666 break; 4667 } 4668 skip: 4669 FILEDESC_SUNLOCK(fdp); 4670 fddrop(fdp); 4671 pddrop(pdp); 4672 free(kif, M_TEMP); 4673 free(okif, M_TEMP); 4674 return (0); 4675 } 4676 4677 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 4678 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 4679 "Process ofiledesc entries"); 4680 #endif /* COMPAT_FREEBSD7 */ 4681 4682 int 4683 vntype_to_kinfo(int vtype) 4684 { 4685 struct { 4686 int vtype; 4687 int kf_vtype; 4688 } vtypes_table[] = { 4689 { VBAD, KF_VTYPE_VBAD }, 4690 { VBLK, KF_VTYPE_VBLK }, 4691 { VCHR, KF_VTYPE_VCHR }, 4692 { VDIR, KF_VTYPE_VDIR }, 4693 { VFIFO, KF_VTYPE_VFIFO }, 4694 { VLNK, KF_VTYPE_VLNK }, 4695 { VNON, KF_VTYPE_VNON }, 4696 { VREG, KF_VTYPE_VREG }, 4697 { VSOCK, KF_VTYPE_VSOCK } 4698 }; 4699 unsigned int i; 4700 4701 /* 4702 * Perform vtype translation. 4703 */ 4704 for (i = 0; i < nitems(vtypes_table); i++) 4705 if (vtypes_table[i].vtype == vtype) 4706 return (vtypes_table[i].kf_vtype); 4707 4708 return (KF_VTYPE_UNKNOWN); 4709 } 4710 4711 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 4712 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 4713 "Process filedesc entries"); 4714 4715 /* 4716 * Store a process current working directory information to sbuf. 4717 * 4718 * Takes a locked proc as argument, and returns with the proc unlocked. 4719 */ 4720 int 4721 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 4722 { 4723 struct pwddesc *pdp; 4724 struct pwd *pwd; 4725 struct export_fd_buf *efbuf; 4726 struct vnode *cdir; 4727 int error; 4728 4729 PROC_LOCK_ASSERT(p, MA_OWNED); 4730 4731 pdp = pdhold(p); 4732 PROC_UNLOCK(p); 4733 if (pdp == NULL) 4734 return (EINVAL); 4735 4736 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 4737 efbuf->fdp = NULL; 4738 efbuf->pdp = pdp; 4739 efbuf->sb = sb; 4740 efbuf->remainder = maxlen; 4741 efbuf->flags = 0; 4742 4743 PWDDESC_XLOCK(pdp); 4744 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp); 4745 cdir = pwd->pwd_cdir; 4746 if (cdir == NULL) { 4747 error = EINVAL; 4748 } else { 4749 vrefact(cdir); 4750 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); 4751 } 4752 PWDDESC_XUNLOCK(pdp); 4753 pddrop(pdp); 4754 free(efbuf, M_TEMP); 4755 return (error); 4756 } 4757 4758 /* 4759 * Get per-process current working directory. 4760 */ 4761 static int 4762 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 4763 { 4764 struct sbuf sb; 4765 struct proc *p; 4766 ssize_t maxlen; 4767 u_int namelen; 4768 int error, error2, *name; 4769 4770 namelen = arg2; 4771 if (namelen != 1) 4772 return (EINVAL); 4773 4774 name = (int *)arg1; 4775 4776 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); 4777 sbuf_clear_flags(&sb, SBUF_INCLUDENUL); 4778 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); 4779 if (error != 0) { 4780 sbuf_delete(&sb); 4781 return (error); 4782 } 4783 maxlen = req->oldptr != NULL ? req->oldlen : -1; 4784 error = kern_proc_cwd_out(p, &sb, maxlen); 4785 error2 = sbuf_finish(&sb); 4786 sbuf_delete(&sb); 4787 return (error != 0 ? error : error2); 4788 } 4789 4790 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, 4791 sysctl_kern_proc_cwd, "Process current working directory"); 4792 4793 #ifdef DDB 4794 /* 4795 * For the purposes of debugging, generate a human-readable string for the 4796 * file type. 4797 */ 4798 static const char * 4799 file_type_to_name(short type) 4800 { 4801 4802 switch (type) { 4803 case 0: 4804 return ("zero"); 4805 case DTYPE_VNODE: 4806 return ("vnode"); 4807 case DTYPE_SOCKET: 4808 return ("socket"); 4809 case DTYPE_PIPE: 4810 return ("pipe"); 4811 case DTYPE_FIFO: 4812 return ("fifo"); 4813 case DTYPE_KQUEUE: 4814 return ("kqueue"); 4815 case DTYPE_CRYPTO: 4816 return ("crypto"); 4817 case DTYPE_MQUEUE: 4818 return ("mqueue"); 4819 case DTYPE_SHM: 4820 return ("shm"); 4821 case DTYPE_SEM: 4822 return ("ksem"); 4823 case DTYPE_PTS: 4824 return ("pts"); 4825 case DTYPE_DEV: 4826 return ("dev"); 4827 case DTYPE_PROCDESC: 4828 return ("proc"); 4829 case DTYPE_EVENTFD: 4830 return ("eventfd"); 4831 case DTYPE_LINUXTFD: 4832 return ("ltimer"); 4833 default: 4834 return ("unkn"); 4835 } 4836 } 4837 4838 /* 4839 * For the purposes of debugging, identify a process (if any, perhaps one of 4840 * many) that references the passed file in its file descriptor array. Return 4841 * NULL if none. 4842 */ 4843 static struct proc * 4844 file_to_first_proc(struct file *fp) 4845 { 4846 struct filedesc *fdp; 4847 struct proc *p; 4848 int n; 4849 4850 FOREACH_PROC_IN_SYSTEM(p) { 4851 if (p->p_state == PRS_NEW) 4852 continue; 4853 fdp = p->p_fd; 4854 if (fdp == NULL) 4855 continue; 4856 for (n = 0; n < fdp->fd_nfiles; n++) { 4857 if (fp == fdp->fd_ofiles[n].fde_file) 4858 return (p); 4859 } 4860 } 4861 return (NULL); 4862 } 4863 4864 static void 4865 db_print_file(struct file *fp, int header) 4866 { 4867 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) 4868 struct proc *p; 4869 4870 if (header) 4871 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", 4872 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", 4873 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", 4874 "FCmd"); 4875 p = file_to_first_proc(fp); 4876 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, 4877 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, 4878 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode, 4879 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 4880 4881 #undef XPTRWIDTH 4882 } 4883 4884 DB_SHOW_COMMAND(file, db_show_file) 4885 { 4886 struct file *fp; 4887 4888 if (!have_addr) { 4889 db_printf("usage: show file <addr>\n"); 4890 return; 4891 } 4892 fp = (struct file *)addr; 4893 db_print_file(fp, 1); 4894 } 4895 4896 DB_SHOW_COMMAND(files, db_show_files) 4897 { 4898 struct filedesc *fdp; 4899 struct file *fp; 4900 struct proc *p; 4901 int header; 4902 int n; 4903 4904 header = 1; 4905 FOREACH_PROC_IN_SYSTEM(p) { 4906 if (p->p_state == PRS_NEW) 4907 continue; 4908 if ((fdp = p->p_fd) == NULL) 4909 continue; 4910 for (n = 0; n < fdp->fd_nfiles; ++n) { 4911 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 4912 continue; 4913 db_print_file(fp, header); 4914 header = 0; 4915 } 4916 } 4917 } 4918 #endif 4919 4920 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 4921 &maxfilesperproc, 0, "Maximum files allowed open per process"); 4922 4923 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 4924 &maxfiles, 0, "Maximum number of files"); 4925 4926 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 4927 &openfiles, 0, "System-wide number of open files"); 4928 4929 /* ARGSUSED*/ 4930 static void 4931 filelistinit(void *dummy) 4932 { 4933 4934 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 4935 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 4936 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), 4937 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 4938 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, 4939 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); 4940 /* 4941 * XXXMJG this is a temporary hack due to boot ordering issues against 4942 * the vnode zone. 4943 */ 4944 vfs_smr = uma_zone_get_smr(pwd_zone); 4945 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 4946 } 4947 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 4948 4949 /*-------------------------------------------------------------------*/ 4950 4951 static int 4952 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 4953 int flags, struct thread *td) 4954 { 4955 4956 return (EBADF); 4957 } 4958 4959 static int 4960 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 4961 struct thread *td) 4962 { 4963 4964 return (EINVAL); 4965 } 4966 4967 static int 4968 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 4969 struct thread *td) 4970 { 4971 4972 return (EBADF); 4973 } 4974 4975 static int 4976 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 4977 struct thread *td) 4978 { 4979 4980 return (0); 4981 } 4982 4983 static int 4984 badfo_kqfilter(struct file *fp, struct knote *kn) 4985 { 4986 4987 return (EBADF); 4988 } 4989 4990 static int 4991 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 4992 { 4993 4994 return (EBADF); 4995 } 4996 4997 static int 4998 badfo_close(struct file *fp, struct thread *td) 4999 { 5000 5001 return (0); 5002 } 5003 5004 static int 5005 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5006 struct thread *td) 5007 { 5008 5009 return (EBADF); 5010 } 5011 5012 static int 5013 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5014 struct thread *td) 5015 { 5016 5017 return (EBADF); 5018 } 5019 5020 static int 5021 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5022 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5023 struct thread *td) 5024 { 5025 5026 return (EBADF); 5027 } 5028 5029 static int 5030 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 5031 { 5032 5033 return (0); 5034 } 5035 5036 struct fileops badfileops = { 5037 .fo_read = badfo_readwrite, 5038 .fo_write = badfo_readwrite, 5039 .fo_truncate = badfo_truncate, 5040 .fo_ioctl = badfo_ioctl, 5041 .fo_poll = badfo_poll, 5042 .fo_kqfilter = badfo_kqfilter, 5043 .fo_stat = badfo_stat, 5044 .fo_close = badfo_close, 5045 .fo_chmod = badfo_chmod, 5046 .fo_chown = badfo_chown, 5047 .fo_sendfile = badfo_sendfile, 5048 .fo_fill_kinfo = badfo_fill_kinfo, 5049 }; 5050 5051 static int 5052 path_poll(struct file *fp, int events, struct ucred *active_cred, 5053 struct thread *td) 5054 { 5055 return (POLLNVAL); 5056 } 5057 5058 static int 5059 path_close(struct file *fp, struct thread *td) 5060 { 5061 MPASS(fp->f_type == DTYPE_VNODE); 5062 fp->f_ops = &badfileops; 5063 vdrop(fp->f_vnode); 5064 return (0); 5065 } 5066 5067 struct fileops path_fileops = { 5068 .fo_read = badfo_readwrite, 5069 .fo_write = badfo_readwrite, 5070 .fo_truncate = badfo_truncate, 5071 .fo_ioctl = badfo_ioctl, 5072 .fo_poll = path_poll, 5073 .fo_kqfilter = vn_kqfilter_opath, 5074 .fo_stat = vn_statfile, 5075 .fo_close = path_close, 5076 .fo_chmod = badfo_chmod, 5077 .fo_chown = badfo_chown, 5078 .fo_sendfile = badfo_sendfile, 5079 .fo_fill_kinfo = vn_fill_kinfo, 5080 .fo_flags = DFLAG_PASSABLE, 5081 }; 5082 5083 int 5084 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, 5085 int flags, struct thread *td) 5086 { 5087 5088 return (EOPNOTSUPP); 5089 } 5090 5091 int 5092 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 5093 struct thread *td) 5094 { 5095 5096 return (EINVAL); 5097 } 5098 5099 int 5100 invfo_ioctl(struct file *fp, u_long com, void *data, 5101 struct ucred *active_cred, struct thread *td) 5102 { 5103 5104 return (ENOTTY); 5105 } 5106 5107 int 5108 invfo_poll(struct file *fp, int events, struct ucred *active_cred, 5109 struct thread *td) 5110 { 5111 5112 return (poll_no_poll(events)); 5113 } 5114 5115 int 5116 invfo_kqfilter(struct file *fp, struct knote *kn) 5117 { 5118 5119 return (EINVAL); 5120 } 5121 5122 int 5123 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 5124 struct thread *td) 5125 { 5126 5127 return (EINVAL); 5128 } 5129 5130 int 5131 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 5132 struct thread *td) 5133 { 5134 5135 return (EINVAL); 5136 } 5137 5138 int 5139 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 5140 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 5141 struct thread *td) 5142 { 5143 5144 return (EINVAL); 5145 } 5146 5147 /*-------------------------------------------------------------------*/ 5148 5149 /* 5150 * File Descriptor pseudo-device driver (/dev/fd/). 5151 * 5152 * Opening minor device N dup()s the file (if any) connected to file 5153 * descriptor N belonging to the calling process. Note that this driver 5154 * consists of only the ``open()'' routine, because all subsequent 5155 * references to this file will be direct to the other driver. 5156 * 5157 * XXX: we could give this one a cloning event handler if necessary. 5158 */ 5159 5160 /* ARGSUSED */ 5161 static int 5162 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 5163 { 5164 5165 /* 5166 * XXX Kludge: set curthread->td_dupfd to contain the value of the 5167 * the file descriptor being sought for duplication. The error 5168 * return ensures that the vnode for this device will be released 5169 * by vn_open. Open will detect this special error and take the 5170 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 5171 * will simply report the error. 5172 */ 5173 td->td_dupfd = dev2unit(dev); 5174 return (ENODEV); 5175 } 5176 5177 static struct cdevsw fildesc_cdevsw = { 5178 .d_version = D_VERSION, 5179 .d_open = fdopen, 5180 .d_name = "FD", 5181 }; 5182 5183 static void 5184 fildesc_drvinit(void *unused) 5185 { 5186 struct cdev *dev; 5187 5188 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 5189 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 5190 make_dev_alias(dev, "stdin"); 5191 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 5192 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 5193 make_dev_alias(dev, "stdout"); 5194 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 5195 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 5196 make_dev_alias(dev, "stderr"); 5197 } 5198 5199 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 5200